Engineering Structures 27 (2005) 1715–1725 www.elsevier.com/locate/engstruct
Technology developments in structural health monitoring of large-scale bridges J.M. Ko∗, Y.Q. Ni Department of Civil and Structural Engineering, The Hong Kong Polytechnic University, Hung Hom, Kowloon, Hong Kong Available online 14 July 2005
Abstract The significance of implementing long-term structural health monitoring systems for large-scale bridges, in order to secure structural and operational safety and issue early warnings on damage or deterioration prior to costly repair or even catastrophic collapse, has been recognized by bridge administrative authorities. Developing a long-term monitoring system for a large-scale bridge—one that is really able to provide information for evaluating structural integrity, durability and reliability throughout the bridge life cycle and ensuring optimal maintenance planning and safe bridge operation—poses technological challenges at different levels, from the selection of proper sensors to the design of a structural health evaluation system. This paper explores recent technology developments in the field of structural health monitoring and their application to large-scale bridge projects. The need for technological fusion from different disciplines, and for a structural health evaluation paradigm that is really able to help prioritize bridge rehabilitation, maintenance and emergency repair, is highlighted. © 2005 Elsevier Ltd. All rights reserved. Keywords: Large-scale bridge; Structural health monitoring (SHM); Instrumentation system; Damage detection; Bridge maintenance
1. Introduction The development of structural health monitoring technology for surveillance, evaluation and assessment of existing or newly built bridges has now attained some degree of maturity. On-structure long-term monitoring systems have been implemented on bridges in Europe [1–4], the United States [5,6], Canada [7,8], Japan [9,10], Korea [11,12], China [13–15] and other countries [16–18]. Bridge structural health monitoring systems are generally envisaged to: (i) validate design assumptions and parameters with the potential benefit of improving design specifications and guidelines for future similar structures; (ii) detect anomalies in loading and response, and possible damage/deterioration at an early stage to ensure structural and operational safety; (iii) provide real-time information for safety assessment immediately after disasters and extreme events; (iv) provide evidence and instruction for planning and prioritizing bridge ∗ Corresponding author. Tel.: +852 2766 5037; fax: +852 2766 1354.
E-mail address:
[email protected] (J.M. Ko). 0141-0296/$ - see front matter © 2005 Elsevier Ltd. All rights reserved. doi:10.1016/j.engstruct.2005.02.021
inspection, rehabilitation, maintenance and repair; (v) monitor repairs and reconstruction with the view of evaluating the effectiveness of maintenance, retrofit and repair works; and (vi) obtain massive amounts of in situ data for leadingedge research in bridge engineering, such as wind- and earthquake-resistant designs, new structural types and smart material applications. The development and implementation of a structural health monitoring system capable of fully achieving the above objectives and benefits is still a challenge at present, and needs well coordinated interdisciplinary research for full adaptation of innovative technologies developed in other disciplines to applications in the civil engineering community. Actually, structural health monitoring has been a subject of major international research in recent years [19–21]. The research in this subject covers sensing, communication, signal processing, data management, system identification, information technology, etc. It requires collaboration between civil, mechanical, electrical and computer engineering among others. The current challenges for bridge structural health monitoring
1716
J.M. Ko, Y.Q. Ni / Engineering Structures 27 (2005) 1715–1725
are being identified as distributed and embedded sensing, data management and storage, data mining and knowledge discovery, diagnostic methods, and presentation of useful and reliable information to bridge owners/managers for decision making on maintenance and management. In this article, after an overview of current status of large-scale bridge health monitoring practice, the authors explore some key issues concerning the above challenges, in a perspective of both researchers and practicers, by referring to several health monitoring engineering paradigms.
improvement from the Tsing Ma Bridge monitoring system to the Stonecutters Bridge monitoring system includes more environment-measuring sensors such as corrosion sensors, barometers, hygrometers and pluviometers to facilitate bridge safety/reliability assessment [40]. Another example is the Sutong Bridge monitoring system that will incorporate a majority of the embedded sensors currently belonging to a foundation stability and safety monitoring system (designed for construction monitoring only) to enhance the superstructure long-term monitoring system for bridge durability assessment.
2. State-of-the-practice in bridge monitoring systems 3. Sensing and data acquisition systems Successful implementation and operation of long-term structural health monitoring systems on bridges have been widely reported. So far about 40 long-span bridges (with spans of 100 m or longer) worldwide have been instrumented with structural health monitoring systems [22]. Typical examples are the Great Belt Bridge in Demark [1], the Confederation Bridge in Canada [23], the Tsing Ma Bridge in Hong Kong [24], the Commodore Barry Bridge in United States [25], the Akashi Kaikyo Bridge in Japan [26], and the Seohae Bridge in Korea [27]. Table 1 lists 20 large-scale bridges in China (including the Hong Kong Special Administrative Region) instrumented with real-time monitoring systems. This listing does not comprise the East Sea Bridge (consisting of two cable-stayed bridges with main spans of 420 m and 332 m respectively), the Hangzhou Bay Bridge (consisting of two cable-stayed bridges with main spans of 448 m and 318 m respectively) and the 3rd Nanjing Yangtze River Bridge (a cable-stayed bridge with a main span of 648 m) of which the long-term structural health monitoring systems are currently under design. Several recent trends in structural health monitoring practice for large-scale bridges are worth mentioning. (i) For some recent bridges such as the Shenzhen Western Corridor, the Stonecutters Bridge, the Shanghai Chongming Crossing (a cable-stayed bridge with a main span of 1200 m) and the Messina Strait Bridge (a suspension bridge with a main span of 3300 m), the design of a monitoring system is required in the tender as part of the bridge design. Integration of bridge design and monitoring system design ensures that design engineers’ important concerns are reflected in the monitoring system while civil provisions for implementing a monitoring system are considered in the bridge design. (ii) The implementation of longterm monitoring systems on new bridges such as the 4th Qianjiang Bridge, the Shenzhen Western Corridor, the Stonecutters Bridge and the Sutong Bridge is accomplished in synchronism with the bridge construction progress. In this way some specific types of sensors, e.g., corrosion sensors, strain gauges and fiber optic sensors can be embedded into the structure during certain bridge erection stages. (iii) The recently devised long-term health monitoring systems emphasize multi-purpose monitoring of the bridge integrity, durability and reliability. In Hong Kong, an
In the past decade significant progress has been made in sensing technology and kinds of innovative sensing systems such as fiber optic sensors and wireless sensors are now becoming commercially available [41–44]. A sensing system is essential to realizing structural health monitoring of bridges. The envisioned future for bridge health monitoring uses an array of inexpensive, spatially distributed, wirelessly powered, wirelessly networked, embedded sensing devices supporting frequent and ondemand acquisition of real-time information about the loading and environmental effects, structural characteristics and responses. Fiber optic sensors have successfully been applied for long-term structural health monitoring of largescale bridges (e.g., [45,46] among others), whereas the application of wireless sensors for bridge monitoring is still in the technology demonstration stage [43]. It is worth mentioning that some conventional sensors become infeasible and impracticable when applied to largescale bridges for long-term monitoring. For example, it is a difficult task to measure the deflection (absolute displacement) of long-span bridges. The traditional displacement transducers can only be used for relative displacement measurement, while laser transducers and total stations have been proven unsuitable for long-term monitoring of longspan bridges. The current solution to this problem is to use a global positioning system (GPS). However, the application of a GPS for bridge monitoring has two limitations: (i) the measurement accuracy of a GPS is not good enough to completely meet bridge health monitoring requirement; and (ii) a GPS does not work well for monitoring the displacement of piers beneath the bridge deck (caused by ships colliding, settlement, etc.). Strain measurement is another issue which is essential for bridge health assessment. There are two types of commonly used strain sensor: electrical resistance strain gauges and vibrating wire strain gauges. Both of them have defects: electrical resistance strain gauges are capable of measuring dynamic strain but possess low zerostability which results in drift of the measurand over time; vibrating wire strain gauges have high zero-stability but can only be used for quasi-static strain measurement. These deficiencies of traditional strain gauges have invoked increasing applications of innovative fiber optic sensors for long-term
J.M. Ko, Y.Q. Ni / Engineering Structures 27 (2005) 1715–1725
1717
Table 1 Major bridges in China instrumented with long-term monitoring systems No. 1
Bridge name
Bridge type
Location
Main span (m)
Sensors installed
Jiangyin Bridge (after upgrade) [28]
suspension
Jiangsu
1385
(1), (2), (3), (4), (5), (6), (9), (10), (13)
2
1st Nanjing Yangtze River Bridge [29]
steel truss
Jiangsu
160
(1), (2), (3), (4), (5), (7), (14)
3
2nd Nanjing Yangtze River Bridge [30]
cable-stayed
Jiangsu
628
(1), (2), (3), (4), (7), (9), (13), (16)
4
Runyang South Bridge [31]
suspension
Jiangsu
1490
5
Runyang North Bridge [31]
cable-stayed
Jiangsu
406
6
Sutong Bridge [32]
cable-stayed
Jiangsu
1088
(1), (2), (3), (4), (6) (1), (2), (3), (4) (1), (2), (3), (4), (5), (6), (7), (8), (9), (10), (11), (16), (18)
7
Tsing Ma Bridge [15]
suspension
Hong Kong
1377
(1), (2), (3), (4), (5), (6), (7), (12), (18)
8
Kap Shui Mun Bridge [15]
cable-stayed
Hong Kong
430
(1), (2), (3), (4), (5), (6), (7), (12), (18)
Ting Kau Bridge [15]
cable-stayed
Hong Kong
475
(1), (2), (3), (4), (5), (6), (7), (12), (18)
10
Shenzhen Western Corridor [15]
cable-stayed
Hong Kong
210
(1), (2), (3), (4), (5), (7), (8), (15), (16), (17), (18)
11
Stonecutters Bridge [15]
cable-stayed
Hong Kong
1018
12
Tongling Yangtze River Bridge [33]
cable-stayed
Anhui
9
432
(1), (2), (3), (4), (5), (6), (7), (8), (9), (10), (11), (15), (16), (17), (18) (1), (2), (4), (11),(13)
13
Wuhu Bridge [34]
cable-stayed
Anhui
312
(2), (3), (4), (5), (10), (12)
14
Humen Bridge [35]
suspension
Guangdong
888
(3), (6), (11), (12)
15
Zhanjiang Bay Bridge [6]
cable-stayed
Guangdong
480
(1), (2), (3), (5), (6), (9), (11), (14), (16)
16
Xupu Bridge [36]
cable-stayed
Shanghai
590
(2), (3), (4), (7), (12)
17
Lupu Bridge [37]
arch
Shanghai
550
(2), (3), (4), (12)
18
Dafosi Bridge [38]
cable-stayed
Chongqing
450
(2), (3), (4), (5), (10), (12)
19
Binzhou Yellow River Bridge [14]
cable-stayed
Shandong
300
(1), (2), (3), (4), (6), (10)
20
4th Qianjiang Bridge [39]
arch
Zhejiang
580
(1), (2), (3), (4), (9),(13)
Note: (1)—anemometers; (2)—temperature sensors; (3)—strain gauges; (4)—accelerometers; (5)—displacement transducers; (6)—global positioning systems; (7)—weigh-in-motion systems; (8)—corrosion sensors; (9)—elasto-magnetic sensors; (10)—optic fiber sensors; (11)—tiltmeters; (12)—level sensors; (13)—total stations; (14)—seismometers; (15)—barometers; (16)—hygrometers; (17)—pluviometers; (18)—video cameras.
monitoring of large-scale bridges. Fig. 1 illustrates such an application where fiber optic sensors are deployed along the deck length of the suspension Jiangyin Bridge for both strain and temperature measurement. The most attractive feature of fiber optic sensors is their capability of distributed sensing and measurement which will result in elaborate condition monitoring for large-scale bridges. The existent main obstacle to wide acceptance of fiber optic sensors for bridge monitoring application is the lack of engineering demonstration of the durability of the sensors in a harsh environment and long-term performance of their attachment to construction materials. Another promising application of fiber optic sensors for cable-supported bridges is the embedment of sensors inside the bridge cables for both temperature and strain measurement. An interdisciplinary research team in Hong Kong Polytechnic University has devised such a fiber optic sensing system for the cable-stayed Sutong Bridge. In this design shown in Fig. 2, seven out of the wires composing the cable cross-section have been replaced by stainless steel tubes for the deployment of fiber optic sensors. Optic fibers in terms of the Brillouin scattering sensors are laid
‘strain-free’ inside each steel tube for distributed temperature measurement along the cable length. The technology of Brillouin-optical time-domain reflectometry (B-OTDR) is used, by which a laser pulse is launched into the optic fiber that serves as the sensing element and the temperature measurement is achieved by combining the scattering information with propagation time of the laser pulses along the fiber. It is noted that seven galvanized wires have been added at the outermost of the cable cross-section to keep the total area of the wires unaltered. Meanwhile, fiber Bragg grating (FBG) sensors are embedded in the cable ends for strain measurement. The strain of the cable near its anchorages is measured with FBG arrays epoxied onto the outside surface of the steel tubes, as shown in Fig. 2. The FBG arrays consist of three FBG strain sensors spaced 2 m apart. The FBGs are sensitive to both strain and temperature. The temperature of the FBGs is obtained from the B-OTDR system, and therefore the strain applied to the FBGs can be determined. Because the FBG arrays are installed along the steel tubes which are used to accommodate B-OTDR for temperature measurement, extra steel tubes and wire area reduction are eschewed.
1718
J.M. Ko, Y.Q. Ni / Engineering Structures 27 (2005) 1715–1725
Fig. 1. Layout of fiber optic sensors on the Jiangyin Bridge.
Fig. 2. Fiber optic sensors embedded inside a bridge cable cross-section.
J.M. Ko, Y.Q. Ni / Engineering Structures 27 (2005) 1715–1725
1719
Fig. 3. Layout of sensors and DAUs on the Sutong Bridge.
On-structure data acquisition units (DAUs) or outstations are indispensable to structural health monitoring systems for long-span bridges. DAUs are assigned at several locations of the bridge to collect the signals from neighboring sensors, digitize the analog signals and transmit the data into a central room outside the bridge. They also have the function of short-term data storage and preliminary signal processing. For large-scale bridges with densely distributed sensors, optimal deployment of DAUs plays a significant role in assuring the quality and fidelity of the acquired data. The number of DAUs required relies on the number of sensor channels in high sampling rate and in low sampling rate. The placement of the DAUs is primarily dependent on the location of analog-type sensors, especially those with lowsensitivity voltage-signal output. The operating environment and allowable maximum distance between the sensors and DAUs have to be considered to eliminate transmission loss/noise and to protect the DAUs from interference. In implementing the structural health monitoring system for the Sutong Bridge, a distributed data acquisition system based on the PXI/SCXI and MXI-3 techniques has been devised to overcome the transmission cabling length limitation while minimizing the number of DAUs. The system is comprised of seven DAUs connected by a FDDI dual-loop fiber optic network to the central room. Each DAU is comprised of a main station (MS) and optional sub-stations (SSs) as illustrated in Fig. 3. The main station consists of a PXI instrument platform, a signal conditioning system and data acquisition modules, while the sub-station (SS) has no PXI instrument platform. Each sub-station is connected to its main station with an MXI-3 interface kit and is remotely controlled by the main station. The PXI instrument platform with a PXI controller is responsible for the operation of the DAU, signal preprocessing and communication with the central room. In the Sutong Bridge, two sub-stations (SS2-1 and SS6-1; refer to Fig. 3), respectively connected to the main stations MS2 and MS6,
are placed inside the two towers (one inside each tower) near the base to collect data from sensors placed below the pylon base level, including those currently belonging to the foundation stability and safety monitoring system (more than 1400 sensors are involved in this system). Making use of the MXI-3 technique, the transmission cabling length limitation can be released without employing additional DAUs. In this design, each DAU is able to support at most eight sub-stations with the aid of MXI-3 extension slots. Lessons learned from the practice on existing bridge monitoring systems tell us that utmost care must be taken with the protection of DAUs. At least two health monitoring systems for the instrumented bridges listed in Table 1 were found with malfunction in all or some DAUs after operation for a few years, due to improper protection. The on-structure DAUs must be designed against a variety of environmental conditions such as temperature, humidity, lightning and electromagnetic interference. The Sutong Bridge project provides an excellent example to demonstrate this issue. Due to limited space of carriageway, five DAUs in the Sutong Bridge have to be placed inside the box girders. Moreover, in order to prevent the steel girders from corrosion, the structural design requires that the humidity inside the sealed box is kept constant by a dehumidification system, and the inner air is prohibited to circulate with the outer air. We have to meet this requirement in designing the protective system for DAUs. A stainless steel cabinet as illustrated in Fig. 4 is designed to house each DAU and to protect it from dust, temperature, humidity, and electromagnetic interference. An air conditioning system is settled inside the cabinet to accommodate severe temperature conditions in the interior of the steel box girders, which may vary from below 0 ◦ C to over 60 ◦ C. To disable the air inside the cabinet from circulating with the air in the interior of the steel box, two holes of 150 mm in diameter are reserved at the lower deck flange in the vicinity of the main station and a refrigerant entry pipe and a refrigerant exit pipe are used to
1720
J.M. Ko, Y.Q. Ni / Engineering Structures 27 (2005) 1715–1725
Fig. 4. Design of a cabinet for housing and protecting DAUs.
circulate the interior air of the cabinet with the air outside the deck box directly. Also, an isolation transformer is adopted at each DAU to protect it from lightning. 4. Effect of environmental factors on measurement data A main distinction of structural health monitoring systems from conventional measurement systems is that the former incorporates damage diagnostic and prognostic algorithms. Extensive research on structural damage identification algorithms has been conducted in the past decades and literature reviews on this subject are available [47–49]. Most widely studied are vibration-based damage detection methods. The vibration-based damage detection methods use measured changes in dynamic features (mainly modal parameters) to evaluate changes in physical properties that may indicate structural damage or degradation. In reality, however, a civil structure is subjected to varying environmental and operational conditions such as traffic, humidity, wind, solar-radiation and, most important, temperature. These environmental effects also cause change in modal parameters which may mask the change caused by structural damage. The evaluation results on the effectiveness of a variety of vibration-based damage detection methods applied to the I40 Bridge indicated that the environmental effects were one of the main pitfalls limiting the practical applicability of modal-based methods [5,50]. For reliability performance of damage detection algorithms, it is of paramount importance to discriminate abnormal changes in dynamic features caused by structural damage from normal changes due to environmental and operational fluctuations, so that neither
will the normal changes raise a false-positive alarm nor will the abnormal changes raise a false-negative alarm in damage detection. With a thorough understanding of the effect of environmental variability on modal properties, it is possible to detect subtle structural damage by incorporating a well defined environmental effect model into appropriate damage detection algorithms [51–54]. Numerous investigations indicate that temperature is the critical source causing modal variability, and the variations of modal frequencies caused by temperature may reach 5% to 10% for highway bridges, which in most cases exceed the changes of frequencies due to structural damage or deterioration. Since long-term structural health monitoring systems for large-scale bridges usually include both vibration transducers and temperature sensors, quantitative understanding and modeling of the effect of temperature on modal properties can be made by using the measurement data covering a full cycle of in-service environmental conditions. With one year of measurement data from the instrumented Ting Kau Bridge [55], a comparative study of evaluating the effectiveness of various statistical regression/learning methods [56] for modeling the effect of temperature on modal frequencies is being conducted by the writers. As part of a long-term structural health monitoring system for this cable-stayed bridge, a total of 83 temperature sensors and 45 accelerometers (67 channels) have been installed for real-time measurement of temperature and dynamic response [15,24]. The modal frequencies of the bridge are obtained by applying an automatic modal identification program to the measured acceleration data at one-hour intervals, and the corresponding temperatures at 20 selected sensor locations are obtained by averaging over
J.M. Ko, Y.Q. Ni / Engineering Structures 27 (2005) 1715–1725
1721
Fig. 5. Frequency sequences measured and generated by the linear regression model: (a) training data; (b) validation data.
Fig. 6. Frequency sequences measured and generated by the nonlinear regression model: (a) training data; (b) validation data.
one hour. A total of 770 hours of data covering one year’s measurement are used for studying the correlation between the modal frequencies and temperatures. The correlation analysis has been conducted by means of linear and nonlinear regression models, neural network models and support vector machine models. Although several investigators [57,58] suggested taking into account thermal inertia by relating current output not only with current input but also with previous input, we use only static correlation models because the measurement data sequence is not always temporally continuous at constant sampling intervals. Fig. 5 shows the first modal frequency sequences obtained by measurement and by the linear regression model. The linear regression model is obtained by least-squares fitting of the training data. It is observed that this model reproduces the training data well but is poor in predicting fresh validation data. Also illustrated in the figure is the distribution of residual error. An adequate model should generate the distribution close to Gaussion population. Fig. 6 shows the corresponding results obtained by a nonlinear ridge regression model in terms of the Radialbasis kernel function [59]. The nonlinear regression model
shows stronger generalization capability (less standard deviation of residual error and better normality of the error distribution) than the linear model, but is still short of accurately predicting the frequency variations. Figs. 7 and 8 illustrate the reproduced and predicted results obtained by neural network and support vector machine models, respectively. These two models exhibit good capabilities in both reproduction and prediction. It is found that a perceptron neural network with a single hidden layer is sufficient for modeling the correlation and an appropriate number of hidden nodes is crucial to achieve superior prediction performance of the trained model (excessive hidden nodes may cause serious attenuation of the prediction capability). When a support vector machine (SVM) is used for modeling the correlation, the prediction capability of the trained model is heavily dependent on the selection of the SVM coefficients [60]. If the SVM coefficients are optimally determined also using the training data like the model parameters, such an obtained model may manifest overfitting and perform poorly in prediction. A more advisable way is to determine the model parameters by means of
1722
J.M. Ko, Y.Q. Ni / Engineering Structures 27 (2005) 1715–1725
Fig. 7. Frequency sequences measured and generated by the neural network model: (a) training data; (b) validation data.
training data while optimizing the SVM coefficients with the use of independent validation data. 5. Linkage with bridge maintenance and management The development of structural health monitoring methods for the detection of damage occurrence, location and severity has now attained some degree of maturity. However, the application of these monitoring data/results for instructing bridge inspection, maintenance and management is still in its infancy [61]. A gap between health monitoring technology and bridge inspection, maintenance and management exercises exists currently which impedes bridge managers from benefiting from the monitoring system. From the monitoring data the bridge managers want to get answers to the serviceability and reliability issues: (a) has the load capacity or resistance of the structure changed? (b) what is the probability of failure of the structural members and the whole structure? Indicators of these performance issues are needed to enable the owners to allocate resources toward inspection, maintenance and rehabilitation of their structures. A method for evaluating the failure probability (or safety index) of bridge components based on long-term monitoring data and time-dependent reliability analysis is
Fig. 8. Frequency sequences measured and generated by the support vector machine model: (a) training data; (b) validation data.
proposed by the writers and envisaged for application in the management of the instrumented Sutong and Jiangyin Bridges. The proposed method is targeted to provide quantitative information to bridge managers for decision making on optimizing and prioritizing bridge inspection and maintenance. According to the reliability theory, the failure probability P f (or safety index β) of a structural component can be evaluated by considering both the member resistance (capacity) R and the load effect S as random variables: f R (r ) f S (s)dr ds (1) Pf = r−s