Structural Health Monitoring and Structural Identification for Long-Span Bridges Kirk A. Grimmelsman Department of Civil Engineering University of Arkansas Fayetteville, USA
[email protected]
Abstract—Structural Health Monitoring and Structural Identification are concepts that are more frequently being applied to improve the reliability and effectiveness of operational and maintenance management for constructed systems. This paper describes why the need for such paradigms continues to grow, and discusses many of the limitations associated with current approaches for assessing and evaluating the performance and condition of civil infrastructure. Several examples of fullscale testing programs applied to long-span bridges are presented to illustrate typical application scenarios and quantitative characterization methods that can be employed. These examples also describe possible sources of uncertainty that can be encountered when executing experimental characterization programs for long span bridges. Keywords-structural health monitoring; constructed systems; experimental characterization; bridges
I.
INTRODUCTION
The nation’s constructed infrastructure systems, which include highly interconnected and interdependent transportation, energy, water, and communications networks, are critical for promoting and sustaining the economic vitality and general societal welfare of the U.S. Given the critical importance of these infrastructure assets and that the present operational performance and condition of many critical infrastructure systems are considered unsatisfactory under normal service demands, there has been increasing interest by researchers, engineers, and infrastructure owners/managers in leveraging technology for preserving, protecting, and managing the nation’s public transportation and other constructed infrastructure systems. This interest has been further stimulated by a number of factors including funding constraints that inevitably require aged and deteriorated structures to remain in service for increasingly longer periods of time with seemingly ever increasing service and performance demands placed on them, a wider recognition of limitations associated with current infrastructure management practices, several well-publicized
failures of constructed systems by accidents, and natural or manmade hazards, and the numerous advances in sensing, data acquisition and associated information and computing technologies that have occurred over the past several decades. A recent report by the American Society of Civil Engineers estimated that $2.2 Trillion would need to be invested over five years to make the condition of the nation’s overall public infrastructure good, and of that total, $930 Billion needs to be devoted to roads and bridges [1]. Since it is unlikely that the required funding levels will ever be readily available for infrastructure investment, there is a compelling need to identify and implement new paradigms that can more efficiently and reliably allocate limited resources where they are needed most. Structural health monitoring has increasingly become viewed as a technology-based paradigm that could lead to more reliable and efficient lifecycle management of constructed systems, especially bridges. This paper will examine some of the challenges and requirements related to structural health monitoring of constructed systems, and in particular long-span bridges. Some of the practical barriers that need to be resolved before this paradigm can be applied on a more routine basis are presented and discussed. Structural identification is an important framework that enables structural health monitoring applications to be designed and executed in a rational and reliable manner. The concepts of structural health monitoring and structural identification both rely on the optimal integration of analytical, experimental and information technologies to characterize the performance and condition of constructed systems. Several examples of full-scale test that were performed to characterize long-span bridges are briefly presented to illustrate how different experimental technologies are applied to such structures. These examples also describe some sources of uncertainty that are typically encountered when attempting to experimentally characterize such structures.
II.
BARRIERS TO STRUCTURAL HEALTH MONITORING OF CONSTRUCTED SYSTEMS
It is notable that despite the many benefits and efficiencies that may be expected from structural health monitoring of constructed systems, and the significant amount of research in this area that has occurred across many industries and engineering disciplines in the past few decades, this paradigm has yet to be widely and routinely implemented in practice for constructed systems. It may be argued that there are a number of important unresolved barriers that have helped to prevent this from occurring. These barriers can be described as those related to technology needs, and those related to fundamental knowledge gaps and uncertainty. The technology related barriers are generally those associated with sensing, data acquisition, computing, communications and information processing and analysis. In the case of constructed systems, the technology related barriers may be considered as less significant since capabilities in this area continue to advance and evolve at a rapid pace. Furthermore, the capabilities of many current sensing and computing technologies have evolved beyond the present ability of engineers to utilize them effectively for structural health monitoring applications. A more significant barrier to routine structural health monitoring of constructed systems are the fundamental knowledge gaps and uncertainty related to the actual behavior of these systems and their many interactions with other systems. The knowledge gaps and uncertainty associated with characterizing constructed systems are distinct from their manufactured systems counterparts. Constructed infrastructure systems are distinguished by the following characteristics that ultimately lead to knowledge gaps and uncertainty related to their actual in-service behavior and performance: (1) the constructed/fabricated nature of these systems can lead to variations in the material properties and mechanical characteristics, and in the types and distribution of any initial defects; (2) the use of different materials and variations in their properties; (3) the typically large physical scale of these systems; (4) differences in structural designs and details which results in many constructed systems being effectively considered as unique systems; (5) variations in the site conditions and subsurface characteristics (geotechnical characteristics, faults, etc.); (6) exposure to uncertain or unknown operating and environmental loads; and (7) very long service lives with direct exposure to harsh environmental conditions. The many unknown or less understood behaviors of constructed systems and their complex interactions with other systems often lead to various forms of load effects and intrinsic responses, and complex aging and deterioration mechanisms that frequently impact their operations, serviceability and durability. The existing knowledge gaps lead to epistemic uncertainty in characterizing and evaluating constructed systems, and must be considered when evaluating the capabilities of various experimental, analytical, and information technologies and the most effective strategies for utilizing them in the context of structural health monitoring for constructed systems.
III.
LIMITATIONS OF CURRENT CONDITION EVALUATION PRACTICES
Presently, the operational and maintenance management of most constructed systems, including bridges, is predominately based on visual inspection data. Visual inspection is subject to a number of limitations that have been shown to affect the reliability of the resulting structural evaluations [2]. In addition to these limitations, visual inspection data is inherently qualitative in nature, and the resulting descriptions of bridge health or performance are qualitative. Even in cases where visual inspection data is used to seed more sophisticated analyses, such as for load rating of bridges, the overall reliability of the results will be limited to the extent that the visual inspection data can reliably describe the condition of the structure. Another important limitation associated with visual inspection is that the data can only describe defect, deterioration and damage conditions at the surface of accessible locations. In the case of large-scale constructed systems, obtaining access to critical locations for visual inspection can by itself represent a significant challenge. Furthermore, any subsurface or hidden defects, deterioration or damage will not be reflected by visual inspection data and the resulting assessment of bridge condition will not be very reliable. Visual inspection also yields very local descriptions of any identified distress. The end user of the data must “connect the dots” between these local descriptions in order to describe the overall or global condition of the structure. Extrapolating from very local descriptions of condition to a global characterization of a structure’s health will amplify any uncertainties that are present in the descriptions of local conditions. Furthermore, visual inspection data can be of limited value for diagnosis and prognosis of certain types of performance problems encountered with operating bridges such as fatigue or corrosion. Finally, condition assessments that rely on visual inspection data as the starting point are not particularly amenable to being performed rapidly. This can be an especially critical issue for recovery operations in the aftermath of accidents or natural hazards such as hurricanes or earthquakes. The predominately subjective data presently used to characterize and evaluate the performance and health of inservice bridges and other constructed systems are clearly not sufficient to enable more proactive, reliable and efficient maintenance management procedures. Furthermore, the limitations associated with the data used for current condition evaluation practices limit the timeliness, reliability and effectiveness of any resulting management decisions [3]. The deficiencies associated with current operational and maintenance management programs may be addressed by quantitatively characterizing the health of bridges and other constructed systems at both global and local levels in terms of minimum performance criteria at each of the operational, serviceability, security and safety limit states. Structural health monitoring is an approach that can enable such quantitative characterizations.
IV.
STRUCTURAL HEALTH MONITORING
Structural health monitoring may be defined in a general sense as measuring, characterizing and tracking a parameter or a spectrum of parameters, in conjunction with analytical simulations and heuristic experience, that quantitatively describe a structure’s current performance or condition. Proactive diagnosis and prognosis of structural performance and condition in terms of rational indices is a necessary precursor for optimal lifecycle management of constructed systems. In addition, structural health monitoring based on rational indices that relate to operational and structural safety may prove invaluable for rapidly evaluating structures in the immediate aftermath of natural or manmade hazard events and for security applications. At a minimum, structural health monitoring requires an effective integration of various analytical, experimental and information technologies. Structural health monitoring also requires that the associated sensing, communications, and computing technologies be designed and employed in the context of a structural identification framework in order to effectively and reliably conceptualize, observe and document the actual mechanical properties, intrinsic forces and distortions, loading and operating environment, and the local and global performance and behavior of constructed systems [4, 5, 6]. Structural identification may be described as a series of six fundamental steps, as shown in Fig. 1, that are executed in conjunction with full-scale testing of constructed systems. The structural identification procedure often must be executed in an iterative manner until an acceptable identification is obtained. The individual steps in this framework also serve to guide the selection, integration, application and interpretation of appropriate experimental, analytical and information technologies for structural health monitoring applications. This framework also helps to minimize or mitigate uncertainty that can occur at the different stages of an experimental characterization program. Finally, the resulting characterization of the structure obtained from structural identification represents a quantitative baseline of performance and condition that can be tracked and evaluated by continuous monitoring, intermittent monitoring, or subsequent full-scale tests. The experimental characterization step is arguably one of the most technology-reliant steps in the overall structural identification framework. The type and scope of the experimental characterization necessary for a reliable structural identification will depend on factors such as the objectives for the characterization, the type and complexity of structure being evaluated, and the age of the structure and its present condition. For instance, if the structural identification is expected to provide a baseline characterization of the structure that will be subsequently utilized for structural health monitoring, the scope and detail of the experimental characterization should be expected to be greater than what would be required for use in designing a structural retrofit. There are many available full-scale testing methods that can be used to characterize a constructed system at either local or global levels for the purpose of structural identification. In the case of bridges, the most commonly utilized experimental approaches, beyond local NDE characterization methods, are
summarized in Fig. 2. These methods can be characterized as either static or dynamic testing methods, in which the latter is concerned with measuring the vibration characteristics of a structure. The individual test methods belonging to each of these two approaches may be further classified as either a controlled or uncontrolled test method, which describes the structural loads (input) being used to conduct the test. A final distinction that can be made between the different testing approaches is whether or not the inputs can be measured and characterized in addition to the structural responses they cause. Ideally, a controlled and measureable input would always be used to characterize constructed systems; however, this is frequently not possible for bridges due to size considerations, or because of prohibitions against disrupting the normal service of the structure. V.
DYNAMIC CHARACTERIZATION EXAMPLE
Dynamic testing is one experimental method that can be used to quantitatively characterize constructed systems at either local or global levels. The resulting characterization can be used to diagnose serviceability problems related to vibration, to calibrate analytical models using the structural identification framework, or as a baseline for structural health monitoring. This characterization approach is frequently employed in practice to develop field-measurement calibrated finite element models of existing long span bridges in conjunction with seismic evaluation and retrofit design programs; however, once the experimental characterization is obtained, it can readily be utilized for structural health monitoring applications.
(1) Conceptualization (6)
(2)
Knowledge & Decisions
A-priori Modeling
Structural Identification (5)
(3)
Model Calibration
Experimental Characterization (4) Data Analysis & Interpretation
Figure 1. Structural identification framework
Static or Quasi-Static Testing
Input Source
Measurands
Controlled
Uncontrolled
Dynamic Testing
Controlled
Uncontrolled
Measure Input & Response
Measure Input & Responses
Measure Responses Only
Measure Inputs & Responses
Measure Responses Only
Static trucks Crawling trucks Special loading devices
Traffic
Traffic
Pull-release devices Impact devices Shaker devices Measured seismic loads
Traffic Wind Waves Pedestrians Un-measured seismic loads Operating machinery or equipment
Figure 2. Full-scale testing methods for structural identification of bridges
In the following example, full-scale dynamic testing was conducted on the Brooklyn Bridge to support a seismic evaluation and retrofit program for the bridge. This landmark suspension bridge spans the East River between the Boroughs of Manhattan and Brooklyn in New York City. The bridge consists of a single 486m-long main span and two 283m-long land spans. The two main towers are massive masonry structures constructed from large granite stones. The towers consist of three columns joined at their tops by Gothic arches above the roadway deck. Each tower is approximately 84m-tall measured from the mean high water elevation, and is supported on large concrete-filled timber caissons. The cross sectional dimensions of the Manhattan tower are slightly larger than those of the Brooklyn tower. The base of the Brooklyn tower has nominal plan dimensions of 43m by 17m, while the Manhattan Tower has nominal plan dimensions of 40m by 14m. The Manhattan tower also weighs more than the Brooklyn tower, with the self-weight for each tower approximately equal to 863MN and 703MN, respectively. The scope of the experimental characterization in this application consisted of identifying the dynamic characteristics of the two masonry bridge towers. The dynamic characteristics of interest were the natural frequencies of vibration, mode shapes and damping ratios. These characteristics, which can be identified from field measurements of the bridge’s vibration responses, are functions of the actual mass and stiffness properties (including the boundary, connectivity, and continuity conditions) of the structure. These same dynamic characteristics and structural system properties can be readily identified from analytical models of the structure. The development of analytical models always includes some degree of idealization. The field-measured dynamic characteristics of the bridge can be used to “tune” or “calibrate” analytical or finite element models of the bridge to more reliably reflect the actual in-service characteristics and behavior. This enables any simulations or analyses conducted using the analytical model to be more realistic and reliable. The dynamic testing involved measuring the vibration responses in vertical, lateral and longitudinal directions at several locations on both the towers and spans. A total of 43 single axis accelerometers were deployed on the structure throughout the dynamic testing program. The accelerometers
were installed at various elevations on the exterior face of the towers (Fig. 5). The measurement data were sampled at either 20 samples per second or at 40 samples per second to compare the effect of different sampling rates on the results. The duration of the testing program was about one month. The measured vibrations were caused by unmeasured and uncontrolled dynamic excitation from the service and natural loads acting on the structure. This specific experimental approach is commonly referred to as “ambient vibration testing,” “output only vibration testing” or “operational modal analysis” since only the dynamic responses of the structure are measured. For the subsequent data analyses, the unmeasured ambient dynamic excitation is assumed to be a broad band, white noise input. The method is frequently employed for dynamic testing of long-span bridges since controlled dynamic excitation testing is generally not feasible for these structures. The dynamic characteristics of the bridge were identified from the measured acceleration data using several different approaches. One basic approach utilized for the data analysis was the peak picking method. The basic premise of this approach is that when a lightly damped structure is subjected to random excitation, the output autospectrum at any response point will reach a maximum at the frequencies where the excitation spectrum peaks or at frequencies where the frequency response function for the structure peaks. In other words, the peaks in the response spectra are assumed to represent either peaks in the excitation spectrum or the normal modes (resonant frequencies) of the structure. The output spectrum may also contain peaks at frequencies corresponding to spurious noise or errors in the measurement signal. There are several parameters and characteristics that may be considered to help distinguish between the output spectral peaks that are due to structural modes and those that are due to peaks in the excitation spectrum or measurement errors. One characteristic is that all points on a structure responding in a lightly damped normal mode of vibration will either be in phase or 180 degrees out of phase with one another, and this will depend only on the shape of the normal mode. At frequencies where a peak in the output spectra is the result of a peak in the excitation spectra, the phase between any pair of outputs will usually be something other than zero or 180 degrees. The phase between any pair of output measurements can be determined from the cross spectrum estimated between them. The magnitude of the cross spectrum estimated between two output measurements will also peak at the locations of the normal mode frequencies. Ordinary coherence functions can also be used to help identify the peaks associated with the normal mode frequencies. The coherence functions tend to peak at the normal modes since the normal modes appear as narrow band peaks in the output spectra and the signal-to-noise ratio in the calculations is maximized at these frequencies. The normal mode shapes associated with the resonant frequencies can be estimated from the responses of sensors distributed on the structure. The mode shapes are actually operating deflection shapes, but the two are assumed to be the same at a natural frequency. The mode shape at each identified resonant frequency can be constructed using the amplitude and phase of sensor location relative to a selected reference:
location. The reference location is simply a measurement location that all other measurement locations are compared to. In practice, the amplitudes of the mode shapes are extracted from each output location, including an output location selected to serve as a reference. The relative difference in the magnitude between each output location and the reference location gives the amplitude of the mode shape at each degree of freedom. The phase information for each measurement degree of freedom is taken from the cross spectra between each output sensor and the reference sensor at the identified natural frequencies. Many of the ambient vibration tests that have been conducted on long-span bridges have used roving instrumentation schemes in which a single stationary reference measurement location is used in conjunction with a limited number of output accelerometers that are gradually roved across the spans of the bridge. In this scenario, the vibration responses at the reference location and usually only a single roving station are simultaneously measured. This permits only one reference location to be used in constructing the mode shapes at any identified peaks in the output spectra. Since the accelerometers installed during the ambient vibration test of the Brooklyn Bridge remained at fixed locations and were simultaneously recorded throughout the field testing program, the vibration responses from each measurement location could be considered as an output channel and also an independent reference channel. In other words, the vibration response from each sensor location could be compared in turn with the measurements from all of the other sensor locations. The resulting “multiple-reference” mode shapes were expected to provide an additional means for evaluating the reliability and consistency of the identified frequencies and mode shapes, and would help to rule out any frequency peaks corresponding to measurement errors, harmonic excitations, or other questionable frequency peaks. A typical frequency autopower spectra for the longitudinal accelerometers at the various levels on the Brooklyn Tower is shown in Fig. 6. There are quite a few peaks visible in the frequency range from DC to 2 Hz. Unit normalized mode shapes of the tower in the longitudinal response direction (parallel to the span length) that correspond to two of the peaks in autopower spectra are shown in Fig. 7. Each plot in Fig. 7 contains 5 separate deflection shapes per identified frequency. Each of these deflection shapes was constructed by comparing the magnitude and phase from the measurements at each level on the tower to the measurements at a selected reference level. Since the measurement data were acquired simultaneously from all tower levels, the data from each level could be used as a separate reference station. The mode shapes shown in Fig. 7 indicate that the peak identified at 1.387 Hz is consistent with the resonant behavior expected at a natural frequency and this frequency was identified as a normal mode for the tower since the shapes were exactly the same no matter which level of the tower was used as the reference location. The peak identified at 0.718 Hz was ruled out as a possible normal mode for the tower since the corresponding deflection shape changes significantly depending on which tower level measurements are selected as the reference location. The peak in the autopower spectra at
this frequency was deemed as likely being the result of the span vibrations acting as dynamic excitation on the tower since the two components are physically coupled through the main cables, the stay cables and the stiffening trusses. It should be noted that when the same data was analyzed using more sophisticated approaches such as the Stochastic Subspace Identification (SSI) method, only a single mode shape was generated for each frequency. As a result, the consistency of the results from each measurement level on the tower was more difficult to evaluate. There were many challenges encountered in the design and execution of the experiment for this dynamic characterization example. Some of these challenges were due to socio-technical constraints associated with field testing of this landmark longspan bridge. The experimental challenges were followed by challenges related to the analysis and interpretation of the experimental results, and these can add significant uncertainty to the resulting experimental characterization. Much of the uncertainty was related to complex static and dynamic interactions between the very stiff towers and the very flexible spans, and due to the complexity of the ambient dynamic excitation acting on different components of the structure. This uncertainty can be difficult to effectively mitigate due to its epistemic nature, and distinguishes experimental characterizations of large scale, in-service constructed systems from mechanical systems that often may be evaluated under controlled conditions in a laboratory, and sometimes by physically separating the various subcomponents from each other. VI.
CONCLUSIONS
The design and execution of reliable Structural Health Monitoring applications requires an optimal integration of analytical, experimental, and information technologies. Structural identification is a rational framework that serves to guide the design and integration of these technologies, and will also help to identify and reduce or mitigate some of the uncertainty associated with characterizing the performance and condition of large scale constructed systems. The quantitative characterization of a structure that results from the implementation of the structural identification framework also provides a necessary rational baseline that can be employed in conjunction with structural health monitoring applications. The full-scale testing examples illustrated some of the application scenarios and experimental methods employed to develop a quantitative characterization of performance or condition of a long-span bridge. Some possible sources of uncertainty related to the design and execution of the experimental program and the analyses and interpretation of the results were also described. In many cases, uncertainty will occur as a result of issues with the quality of the measurement data. These issues are often difficult to avoid when testing very large in-service structures, even when the design and execution of the field test is performed with great care by experienced personnel. Fortunately, this uncertainty can often be reduced or mitigated through modifications to the experimental setup during the testing program or through the use of different signal processing approaches during the data analysis phase. In other cases, the uncertainty was encountered while analyzing and
interpreting the measurement results. The latter form is generally more difficult to anticipate and therefore mitigate. Often the only effective way to deal with such uncertainty is to conduct the characterization program using the structural identification framework.
A critical step in designing and executing any field experiment or even a single measurement on a constructed system is to first recognize the many possible forms of uncertainty inherent in such efforts, and carefully evaluate if the uncertainty can be sufficiently mitigated. Unfortunately, too many field tests are performed with inadequate measurements and without due attention to human errors and inexperience that may render the results from the experiment more uncertain than the application which they were intended to address. Structural identification and sensitivity analyses are concepts that, if properly applied, may help to reduce or mitigate the impacts of uncertainty on the experimental characterization results.
x 10
-7
B C E G H MT
4
3.5
Acceleration (g 2/Hz)
3
2.5
2
1.5
1
0.5
0
0
0.5
1
1.5
80.0
L
2.5 Frequency (Hz)
Frequency = 0.718 Hz
80.0
70.0
70.0
60.0
60.0
50.0
50.0
40.0
40.0
30.0
30.0
20.0
Level Level Level Level Level
10.0
L
2
T
12m
0.0
L
T
L
T
T
L
L
T
L
T
4
4.5
5
0.2
0.4
0.6
0.8
Frequency = 1.387 Hz
Reference Location Level Level Level Level Level
10.0
B C E G H
0.0 1.0
0.0
0.2
0.4
0.6
0.8
1.0
Amplitude
Amplitude
Figure 5. Tower longitudinal mode shapes corresponding to peaks identified in the autopower spectra 17m
L
3.5
20.0
B C E G H
0.0
T
3
Figure 4. Autopower spectra for tower longitudinal vibrations
Height Above Tower Base (m)
In addition to technological barriers, fundamental gaps remain in understanding how constructed systems behave and many sources of uncertainty affecting full-scale testing applications represent significant barriers to more widespread applications of smart systems paradigms such as structural health monitoring of constructed infrastructures. Uncertainty can affect the design, execution, analysis, and interpretation phases of a field experiment and additional complexity may be introduced by uncertainties inherent in designing, constructing, measuring and simulating large-scale, constructed infrastructure systems. In most cases, the results obtained from the most carefully designed and executed field experiments will still contain some degree of uncertainty which may or may not be possible to completely characterize or mitigate.
4.5
REFERENCES 84m
17m
[1]
L
T
12m
L
[2]
L
T
L
L
T
[4]
16m
10m
L
[3]
[5]
[6] East Elevation Brooklyn Tower (L = Longitudinal Accelerometer, T = Transverse Accelerometer)
Figure 3. Accelerometer locations on the Brooklyn Tower
American Society of Civil Engineers, 2009 Report Card for America’s Infrastructure, Reston, VA, March 2009. M. Moore, B. Phares, B. Graybeal, D. Rolander and G. Washer, Reliability of Visual Inspection for Highway Bridges, Report FHWARD-01-020, US Department of Transportation, FHWA, June 2001. Chase, S.B., “Hi-tech Inspection,” Civil Engineering, Vol. 71, No. 9, ASCE, pp. 62-65, September 2001. A.E. Aktan, D.N. Farhey, A.J. Helmicki, D.L. Brown, V.J. Hunt and A. Levi, “Structural identification for condition assessment: experimental arts,” J. of Struct. Engrg, vol. 123, ASCE, pp. 1674-1684, December, 1997. A.E. Aktan, N. Catbas, A. Turer and Z. Zhang, “Structural identification: analytical aspects,” J.of Struct. Engrg., vol. 124, ASCE, pp 817-829, July, 1998. A.E. Aktan, K.A. Grimmelsman, R.A. Barrish, F.N. Catbas and C.J. Tsikos, “Structural identification of a long span truss bridge,” Trans. Res. Record, vol. 1696, National Academy Press, pp. 210-218, 2000.
