U.S. Department of Transportation
Federal Highway Administration
1200 New Jersey Avenue, SE
Washington, DC 20590
202-366-4000


Skip to content
Facebook iconYouTube iconTwitter iconFlickr iconLinkedInInstagram

Federal Highway Administration Research and Technology
Coordinating, Developing, and Delivering Highway Transportation Innovations

Report
This report is an archived publication and may contain dated technical, contact, and link information

Publication Number: FHWA-RD-03-089
Date: September 2005

Dynamic Bridge Substructure Evaluation and Monitoring

Previous | Table of Contents | Next

Chapter 2. Project Scope and Literature Review

Background

This FHWA research project, "Dynamic Bridge Substructure Evaluation and Monitoring" was proposed and funded in 1995 to investigate the possibility that by measuring and modeling the dynamic response characteristics of a bridge substructure, one could determine the condition and safety of the substructure and identify its foundation type (shallow or deep). Determination of bridge foundation conditions may be applied to quantify losses in foundation stiffness caused by earthquake, scour, and impact events. Identification of bridge foundation type may be employed to estimate bridge stability and vulnerability under dead and live load ratings, particularly for unknown bridge foundations. The dynamic evaluation and monitoring results eventually may be integrated into current and proposed bridge management system (BMS) databases to provide baseline data for comparison of bridge substructures after catastrophic events.

Accurate information on a bridge substructure is an essential part of implementing a cost effective and safety-conscious bridge management program. For bridge piers and abutments, scour can quickly reduce the load capacities of foundations, and ground motions induced by earthquakes can produce ground-coupled resonance, liquefaction of soils, and loss of soil support. Therefore, it is very important that bridge foundation vulnerability to seismic and scour events be assessed or monitored, or both, cost-effectively so that catastrophic failures an be prevented and corrective repairs can be carried out in a timely fashion. Furthermore, more than 100,000 bridges over water in the United States have what is known as "unknown foundations,"(1) which means their vulnerability to scour cannot be calculated by normal hydraulic and geotechnical analysis procedures.

As a potential approach to address this problem of foundation condition and type, dynamic testing is a powerful tool for bridge superstructure and substructure condition assessment and system identification in both global and local element levels. A moderate amount of work has been done over the past 25 years in bridge vibration measurements and comparisons of dynamic field data and analytical predictions, as well as in an attempt to relate dynamic testing results to potential deterioration of a bridge.(2) However, most of the work emphasized dynamic testing and analysis of bridge superstructures with little or no consideration given to bridge substructures and their foundations.(3) The National Cooperative Highway Research Program (NCHRP) Project 21-5 demonstrated promise for a complete bridge substructure nondestructive evaluation (NDE) method based on dynamic testing of bridge substructures.(4) However, funds were limited for research on this new method in the NCHRP research on NDE methods to determine unknown bridge foundation depths for bridge scour safety analyses; thus, only the initial feasibility of the approach could be evaluated before this report.

Research Scope

The objectives of this project were to backcalculate bridge foundation dynamic stiffness and identify bridge foundation type using dynamic measurements of a bridge substructure, and to monitor bridge substructure conditions over time by repeating dynamic tests at preselected time intervals or after the occurrence of a natural event such as an earthquake or flood. The project objectives were met by performing field vibration tests before and after simulated earthquake and scour damage to actual bridge foundations, modal analysis, finite element modeling (FEM), structural parameter estimation, and a new dynamic response system identification technique known as the HHT.(5)

Chapter 3 discusses the details of four bridge bents in three bridges with different foundation conditions and types that were field tested in Texas during 1996 and 1997. All four bridge bents were first tested in their existing, undamaged states. Because one of the bridges was demolished during the field research, two bents could be tested in an undamaged state, and then in a simulated scour state, and finally in a simulated earthquake damage state. The simulated foundation damage conditions were meant to represent the effects of scour and earthquake events on bridge foundations, and the selected foundation types are representative of typical highway bridge foundation configurations in the United States.

Chapter 4 discusses the dynamic testing of the four bridge substructures including instrumentation, test procedures, and data processing used in the field modal vibration tests of the four bridge bents. A specially modified large, truck-mounted geophysical vibrator (a Vibroseis) owned by the Geotechnical Engineering Center of the University of Texas at Austin was used primarily as the vibration source.(6) The Vibroseis consists of a large truck with a gross weight of about 22,246 kilograms (kg) (49,000 pounds (lb)) with a servo-hydraulic vibrator mounted on it. Vertical dynamic forces up to 31,780 kg (70,000 lb) over a frequency range from 1 to 120 hertz (Hz) may be produced with the Vibroseis. Seismic accelerometers were attached to various locations on the tested bridge bents to record bridge dynamic responses under forced vibrations. A personal computer (PC)-based four-channel dynamic signal analyzer served as the central vibration control and measurement unit.

Chapter 5 presents the experimental findings in terms of foundation condition and type identification. Dynamic characteristics of each bridge substructure such as natural frequency, damping ratio, and mode shape were extracted from the field data and interpreted by modal analysis techniques.

Chapter 6 discusses the results of modeling and structural parameter estimation and system identification techniques that were used to produce engineering information such as lateral, vertical, and rotational resistance of the bridge foundations. In terms of theoretical analysis, three-dimensional (3-D) FEMs for all four tested bridge bents were accomplished by using a commercial software package. The 3-D FEMs consist of super-soil-structural elements that can represent the bridge foundation conditions and types. Based on the 3-D FEMs, two-dimensional (2-D) FEMs for the four tested bridge bents were established because the current structural parameter estimation and system identification program used in this project is available for 2-D models only.

Chapter 7 discusses the HHT and results of its application to the two bents that were tested in undamaged, simulated flood, and simulated earthquake states. The HHT method is designed to identify instantaneously at points in time the reduction in natural frequency associated with local damage from otherwise linearly elastic response data for the rest of a structure that are often unidentifiable in conventional modal testing and analysis.

Chapter 8 explores the possibility of incorporating the dynamic bridge substructure testing and analysis results into BMS. This step will be taken in the future to provide practical benefits to departments of transportation (DOT) and the public with research in improved safety and reduced risk of bridge substructure failure.

Chapter 9 presents the findings and conclusions resulting from this research. Recommendations for future research are also discussed.

Project Background and Review

The highway transportation system is the largest and perhaps most important subsystem in the transportation infrastructure of the United States. It helps sustain commerce in almost every sector of our national economy and is used daily for both pleasure and necessity by almost every citizen. Maintaining this system at a high performance level is vital for public safety, societal well-being, and economic productivity and growth. Bridges comprise significant and critical discrete links in the highway transportation subsystem. An estimate made in 1995 was that about 45 percent of the bridge inventory was deficient due to either structural or traffic inadequacy, or both.(7) In 2004, when measured by deck area adjusted for average daily traffic, 29.8 percent of national highway system bridges and 31.5 percent of non-national highway system bridges were deficient.(8)

Clearly, the job of managing and maintaining the transportation infrastructure represents a significant challenge, especially considering the difficulties in generating the level of additional funds necessary to finance these activities. At present, the principal information available to bridge owners comes from FHWA’s National Bridge Inspection Program, which was put in place in 1967 in response to the collapse of the Silver Bridge spanning the Ohio River between Gallipolis, OH, and Point Pleasant, WV. In the intervening 2 decades, many State DOTs initiated and accomplished significant improvements in inspection, rating, and management operations of highway bridges; however, according to FHWA, a fundamental weakness in bridge management systems has been the reliance on visual inspection and subjective condition assessment.(7) Thus, these inspections cannot evaluate damage in the absence of visible symptoms such as concrete deck deterioration under asphalt overlay, corrosion of reinforcement, and scour. Moreover, data analysis is based on characterizations that typically do not incorporate the mechanisms significantly influencing actual bridge behavior. Current practices in evaluating serviceability, fatigue, and ultimate limit capacity of a bridge are accomplished only with significant uncertainty; thus, their effectiveness in management decisionmaking is greatly reduced. Without more accurate information on bridge superstructure and substructure, bridge owners cannot decide where, when, and how to spend limited resources efficiently.

The objective of this research project is to produce a dynamic bridge substructure evaluation and monitoring system that can provide more accurate information on the condition, type, and vulnerability of bridge piers and abutments that cannot be obtained by normal visual surveys. Dynamic testing plays a key role in this bridge substructure evaluation and monitoring system.

Although many researchers have studied dynamic behavior of highway bridges, it is still difficult to understand fully all parameters that affect vibration measurements. Cantieni explained the difficulties encountered when he conducted an experiment on a bridge to study the dynamic behavior of highway bridges under the passage of heavy vehicles.(9) During the literature search, the current researchers realized that most experiments and theoretical studies previously performed were related to dynamic behavior of a bridge superstructure, and little had been done on dynamic behavior of a bridge substructure. The following sections summarize the key issues for bridge substructure condition assessment and relate this assessment to the proposed bridge evaluation and monitoring system.

Damage Risks to Bridge Substructures

Typically, a bridge abutment is designed to resist lateral movement and overturning created by soil pressure and settlement resulting from dead and live loads. The bridge abutment and its connection to the footing must resist moments and shear forces, and the footing must provide resistance to vertical, lateral, and overturning forces. Live loads add slightly to the vertical dead loads, but they also add to the resistance to overturning and sliding. Therefore, the bridge superstructure usually controls the load ratings. A bridge abutment condition rating is governed by three factors: (1) the presence of excessive soil pressure caused by poor drainage, (2) the condition of the abutment structure, and (3) the dimension and type of foundation (shallow, deep, or combined such as footing, pile, or pile cap on pile).

A bridge pier is designed to resist vertical settlement resulting from dead and live loads, and lateral movement and rotation caused by temperature change, friction, wind, water, and seismic loads. The bridge pier and its connection to the footing must resist moments, shear, and compressive forces. The footing must resist lateral, vertical, and rotational movements. The bridge pier condition rating is governed by the condition of the pier structure and the dimension and type of footing (shallow, deep, or combined).

The most common cause of bridge failure is from floods when scour causes failure of bridge piers and abutments.(10) Scour occurs progressively as supporting material under a footing is removed during flood events and is replaced with material that has little or no bearing capacity. During the spring floods of 1987, 17 bridges in New York and New England were damaged and destroyed by flood scour action. In 1985, floods destroyed 73 bridges in Pennsylvania, Virginia, and West Virginia. A national study of 383 bridge failures caused by catastrophic floods showed that 25 percent involved pier damage and 72 percent involved abutment damage.(11) More than 85,000 bridges in the United States are vulnerable to scour; bridge foundation conditions for another 104,000 cannot be determined.(1) Devices have been developed for monitoring scour events.(12,13) These devices could be tied in with the proposed dynamic substructure monitoring system to correlate the bridge pier and foundation response with the potential occurrence of scour.

Substructure damage during earthquakes generally consists of foundation elements broken in shear, or loss of soil support, or both, caused by liquefaction. Rapid, nondestructive identification of such hidden substructure damage after an earthquake would increase public safety.

Barge and ship collisions with bridges are common worldwide, and can represent a significant cause of damage to bridge substructure.(14) From 1981 to 1990, 2,418 bridges in the United States alone were hit by commercial marine vessels. Most did not involve fatalities but may have resulted in damage that could have been monitored and reported.(15) Although damage to substructures can be similar to that from an earthquake, impact damage primarily affects the bridge substructure and there is a comparatively low risk of reduced soil support.

Loss of capacity in bridge piers and abutments can occur either over a period of time resulting from alkali-silica reaction, freeze-thaw damage, corrosion of reinforcement, and unconstrained thermal movements, or because of sudden floods, earthquake, or vessel impact.(16) The long-term damage conditions can be evaluated using a variety of local tests including impact-echo, groundpenetrating radar, and corrosion-potential measurements. However, damage caused by floods, earthquake, and vessel impact is more difficult to evaluate locally for buried portions of bridge substructures, and thus global bridge evaluation and monitoring methods are more suitable (as was researched in this project).

Related Work for Bridge Substructure Condition Evaluation

Typically, a bridge abutment is designed to resist lateral movement and overturning created by soil pressure and settlement resulting from dead and live loads. The bridge abutment and its connection to the footing must resist moments and shear forces, and the footing must provide resistance to vertical, lateral, and overturning forces. Live loads add slightly to the vertical dead loads, but they also add to the resistance to overturning and sliding. Therefore, the bridge superstructure usually controls the load ratings. A bridge abutment condition rating is governed by three factors: (1) the presence of excessive soil pressure caused by poor drainage, (2) the condition of the abutment structure, and (3) the dimension and type of foundation (shallow, deep, or combined such as footing, pile, or pile cap on pile).

A bridge pier is designed to resist vertical settlement resulting from dead and live loads, and lateral movement and rotation caused by temperature change, friction, wind, water, and seismic loads. The bridge pier and its connection to the footing must resist moments, shear, and compressive forces. The footing must resist lateral, vertical, and rotational movements. The bridge pier condition rating is governed by the condition of the pier structure and the dimension and type of footing (shallow, deep, or combined).

The most common cause of bridge failure is from floods when scour causes failure of bridge piers and abutments.(10) Scour occurs progressively as supporting material under a footing is removed during flood events and is replaced with material that has little or no bearing capacity. During the spring floods of 1987, 17 bridges in New York and New England were damaged and destroyed by flood scour action. In 1985, floods destroyed 73 bridges in Pennsylvania, Virginia, and West Virginia. A national study of 383 bridge failures caused by catastrophic floods showed that 25 percent involved pier damage and 72 percent involved abutment damage.(11) More than 85,000 bridges in the United States are vulnerable to scour; bridge foundation conditions for another 104,000 cannot be determined.(1) Devices have been developed for monitoring scour events.(12,13) These devices could be tied in with the proposed dynamic substructure monitoring system to correlate the bridge pier and foundation response with the potential occurrence of scour.

Substructure damage during earthquakes generally consists of foundation elements broken in shear, or loss of soil support, or both, caused by liquefaction. Rapid, nondestructive identification of such hidden substructure damage after an earthquake would increase public safety.

Barge and ship collisions with bridges are common worldwide, and can represent a significant cause of damage to bridge substructure.(14) From 1981 to 1990, 2,418 bridges in the United States alone were hit by commercial marine vessels. Most did not involve fatalities but may have resulted in damage that could have been monitored and reported.(15) Although damage to substructures can be similar to that from an earthquake, impact damage primarily affects the bridge substructure and there is a comparatively low risk of reduced soil support.

Loss of capacity in bridge piers and abutments can occur either over a period of time resulting from alkali-silica reaction, freeze-thaw damage, corrosion of reinforcement, and unconstrained thermal movements, or because of sudden floods, earthquake, or vessel impact.(16) The long-term damage conditions can be evaluated using a variety of local tests including impact-echo, groundpenetrating radar, and corrosion-potential measurements. However, damage caused by floods, earthquake, and vessel impact is more difficult to evaluate locally for buried portions of bridge substructures, and thus global bridge evaluation and monitoring methods are more suitable (as was researched in this project).

Related Work for Bridge Substructure Condition Evaluation

Raghavendrachar and Aktan successfully conducted a multireference impact testing on a reinforced concrete slab bridge.(17) This pilot study demonstrated that a multireference impact testing could serve as the main experimental component for comprehensive structural identification of large constructed facilities. If an accurate measure of flexibility (displacement divided by force as a function of frequency, that is, the inverse of stiffness) is to be obtained directly from the experimental data, demanding standards are required for modal testing designs. Aktan and Helmicki performed a study to explore the issues and advancement of knowledge in instrumented monitoring of a full-scale bridge.(18) When impact testing may not be the appropriate method (such as for buildings, large bridges, large facilities with complex geometry, or structures subjected to lateral loads), forced-excitation modal testing using larger vibrators may be required.(19)

Warren and Malvar used a falling weight deflectometer to assess structural conditions of reinforced concrete piers.(20) By comparing the deflected shapes from the FEM and testing results, the local stiffness and soft areas of the piers were determined. The FEMs of the piers were generated from the design data and drawings using a commercial software package called Automatic Dynamic Incremental Nonlinear Analysis (ADINATM). The differences between a rating of ideal and actual testing data were resolved by the systematic changes of the stiffness parameters based on matching dynamic responses of the FEMs to the measured data. This method was demonstrated on a real bridge with timber piles in New Jersey. Because the bridge geometry was well known, it was deemed sufficient to identify the effective stiffness of the piers and damaged areas.

To investigate the seismic vulnerability of bridge piers, the Washington State DOT studied lateral load responses of a full-scale reinforced concrete bridge.(21) This study addressed concerns that the design of pier columns in the 1950s and 1960s was not adequate to sustain displacements during earthquakes. A three-span highway bridge was pulled transversely using a jacking arrangement and cables that produced a load equal to 65 percent of the weight of the bridge. The overall movement was 7.62 millimeters (mm) (0.3 inches), and very little damage was detected. This study concluded that the pier design was adequate for earthquakes.

Pierce and Dowding reported on a long-term monitoring method for concrete bridge piers using Time Domain Reflectometry (TDR).(22) This method focused on the determination of internal cracking and large local deformations caused by earthquakes. To use the method, coaxial TDR cables must be embedded in the concrete during construction or retrofitting. The cables were placed in critical areas such as the column/base connection. The cables were selected as either extension sensitive or shear sensitive. If localized extension or shear occurred along the length of any cable, it could be identified as a reflector using the TDR external electronics.

A bridge abutment generally is designed to resist backfill soil pressures; however, for a rigid frame abutment, the thermal deck expansion causes backfill pressures that are far in excess of the active soil pressures used in design.(23) In addition, bridge skew results in a large horizontal gradient of the backfill pressures, producing local backfill pressures that could exceed the capacity of the abutment walls. A software program, BASSIN, was developed in 1996 at the University of California, Berkeley, CA, for dynamic analysis of a bridge-abutment-backfill system that is subjected to traveling seismic waves. BASSIN can compute 3-D dynamic responses of an arbitrarily configured bridge-abutment-backfill system induced by compressional, vertical shear, horizontal shear, and surface waves (planar P-, SV-, SH- or Rwaves, respectively) with arbitrary wavelength, amplitude, and direction of incidence.

During the North American Workshop on Instrumentation and Vibration Analysis of Highway Bridges in 1995, researchers and practitioners agreed that instrumentation is a viable tool for bridge inspection. In fact, many State DOTs (such as Connecticut, Florida, and New York) have used instrumentation in their bridge inspection programs. The California DOT (Caltrans) has an extensive instrumentation program that involves monitoring seismic excitations and foundation systems. Practicing bridge engineers recognized the need to evaluate and formalize the use of structural identification and instrumentation for bridge inspection.

Hussein et al. reported the use of compression waves for investigating single pile length and integrity, settlement, and scour.(24) Finno and Prommer studied the impulse response (IR) method for inaccessible drilled shafts under pile caps.(25) Several drilled shafts connected together with concrete grade beams were tested using the nondestructive IR method. Based on the field data, it was found that shaft heads that were more rigid (because of larger or several grade-beam connections) exhibited greater signal attenuation, and become more difficult to evaluate. Chen and Kim used transverse waves as a means for investigating pier conditions and local defects.(26) This so-called "bending wave" method involved the measurements of the velocity dispersion curve of the transverse waves propagated down from the top of a pier. The dispersion of the directly arrived wave was used to assess local damages, while the dispersion of the reflected waves from the pier bottom was used to assess overall pier conditions. The method proved most suitable to short piles in softer soils.(27)

The previously mentioned NCHRP 21-5 research study for determination of unknown subsurface bridge foundations was conceived to evaluate and develop NDE methods and equipment that allowed the determination of subsurface bridge foundation depths and other characteristics where such information is unavailable, unknown, or uncertain.(4) Out of approximately 580,000 highway bridges in the National Bridge Inventory, a large number of older non-Federal-aid bridges, and, to a lesser extent, Federal-aid bridges have no design or as-built bridge plans. Consequently, little or no information is available to document the type, depth, geometry, or materials incorporated in their foundations. The study evaluated many existing and new NDE methods including five acoustic methods (sonic echo and impulse response, bending wave, ultraseismic, parallel seismic, and borehole sonic), one modal vibration method (dynamic foundation response), and one electromagnetic (borehole and surface ground penetrating radar) method. A follow-on phase II study focused on researching and developing equipment, field techniques, and analysis methods for the surface-based ultraseismic and borehole-based parallel seismic methods.(27) These two methods showed the most promise for immediate application to the determination of unknown foundation depths for the most substructures.

Recent research was done on Interstate 15 (I-15) bridges in the Salt Lake City, UT, area on dynamic testing for condition assessment of bridge bents.(28) The research used seven forced vibration tests with horizontal excitation. Modeling and experimental modal vibration test results were compared in terms of mode shapes and frequencies. Both damaged and repaired substructure states could be used to identify the condition of the structure at each state. The estimated location and intensity of the damage or retrofit also was identified.

Related Work for Bridge Superstructure Condition Evaluation

Bridge superstructure condition evaluation research programs generally have focused on two primary areas: ultimate load tests and dynamic tests. An excellent reference on dynamic testing for modal vibration measurement and analysis is given by Ewins.(29) For the ultimate load tests, the bridges that were slated for removal from service were tested to failure.(30,31,32) These studies generally provided some insight on the ultimate load capacity and mechanisms of failure that could be used in the future. On the other hand, laboratory and field studies to evaluate dynamic properties of bridges and relate them to condition assessments have been reported extensively in past years.(33)

Salane et al. reported dynamic tests of a bridge for detecting structural deterioration caused by girder fatigue cracks.(34) A concrete deck on steel girders was loaded with an electrohydraulic actuator system up to 465,000 load cycles. Accelerometers were used to determine damping ratios, frequency contents, and impedance at various stages during the loading. The testing results indicated increases in damping ratios with cycles of loading, presumably caused by cracking, and a decrease in amplitude at resonant frequencies, as well as a 20 percent to 40 percent change in computed stiffness coefficients.

Cawley and Adams related changes of successive mode frequencies to the existence and location of structural deterioration in beams.(35) Manning, DeWolf et al., Huston et al., and Gregory et al. reported various full-scale bridge dynamic tests, showing that dynamic characteristics may be revealed using vibratory shakers, impact hammers, and traffic and wind loads. (See references 36, 37, 38, and 39.) Gregory et al. and DeWolf et al. demonstrated a relationship between dynamic testing results and structural deterioration.(39,37) Sensitivity of dynamic characteristics to deterioration was shown to depend on the particular modes being observed. Manning suggested that a more localized dynamic analysis might be advantageous because serious loss of strength of a single member may occur before it can be observed on the entire structure.

Mazurek and DeWolf showed in field tests that ambient traffic loads could be used as a basis for an automated monitoring scheme based on changes in vibration signatures.(40) They also showed in laboratory tests that changes in support condition and crack development affect natural frequencies and modal amplitudes. Changes in modal frequency were up to 30 percent for changes in support condition and up to 10 percent for cracking. These laboratory results encouraged further field investigations.

Biswas et al. reported a component evaluation technique based on dynamic responses to a hammer impact on the component of interest.(41) Results were confirmed with laboratory models, but field verification was limited and did not produce conclusive results.

Woodward et al. conducted dynamic tests for a full-scale bridge subject to artificially induced fatigue cracking (vertical cuts) in a main girder.(32) Preliminary field test results showed that the changes in dynamic characteristics due to the damage were detected, but only when the maximum amount of damage was inflicted. Salawu and Williams reported a study of the forced vibrations of a bridge before and after repair.(42) The test results demonstrated the changes in natural frequency induced by the repair; however, the magnitude of the changes is quite small, on the order of 2 percent.

The principle of continuous monitoring of bridge dynamic characteristics has been implemented experimentally by the New York State DOT.(43) The system, called the Remote Bridge Monitoring System, is based on measuring dynamic motion (using accelerometers) as well as strain and rotation (using inclinometers). The sensors are hardwired to an onsite data acquisition system, and the acquired data are transmitted to the central office by telephone wire and modem. A threshold accelerometer signal level is used to trigger data collection and a warning alarm to announce significant changes in modal frequencies. Dynamic response data collected to date show some scatter in the modal frequency measurement, in which significant changes on the order of 5 percent to 10 percent might be obscured.

It is possible to conclude from the above studies that further work is needed to relate dynamic properties to component deterioration; however, simpler interpretations for vibration measurements have not been reported in the literature. For example, dynamic measurements can be used to evaluate the distribution of loading in axially loaded members such as cables and truss diagonal braces. The natural frequency of such axially loaded members is highly sensitive to the magnitude of the axial load. Baumgartner and Waubke showed how frequency measurements in tension hangers under traffic loading relate to the end fixity of the hangers.(44)

Bridge Structural Parameter Identification Work

The objective of structural parameter identification is to obtain an understanding of the critical mechanisms of flexibility, energy dissipation, and inertia. 3-D kinematics, resistance mechanisms (load paths), and critical structure region studies (with respect to stress, strength, and stiffness) are essential to reliably assess and identify the available supplies of strength, stiffness, stability, hardening, and energy dissipation in a structure. Structural parameter identification consists of two subactivities-experimental testing results and parameter identification.

Structural parameter identification is the art of reconciling an analytical model of a structure with experimental data using optimization. The identified parameters should be useful for structural condition assessment in the geometric space. The principle behind structural parameter identification is not new, but, in recent years, increased computational capability has resulted in significant progress in algorithm development and experimental data processing. (See references 45, 46, 47, 48, and 49.) Doebling et al. summarized current damage identification methods using measured vibration data; however, none of the papers cited discussed parameter identification or bridge foundation condition assessment.(50)

As a minimum input, structural parameter identifications require some structural response measurements to static or dynamic loads. Figure 1 shows typical excitations and outputs of a parameter identification system. Structural responses to controlled or operating excitations are combined within the parameter identification system to produce either structural modal estimates or structural element parameter estimates. When controlled excitations are used, the excitation information can be added in the parameter identifications; however, a more practical approach is to perform structural parameter identifications with unknown operating excitations.

 

View Alternative text

Figure 1. Diagram. Parameter identification system excitation and output options.

 

Figure 2 shows that structural parameter identification can be classified into problems with and without mathematical models. Then the mathematical models are subdivided into static and dynamic models. The static model is further divided into displacement and rotational measurements and strain measurements. The dynamic model is divided into time domain and frequency domain. Structural parameter identification with no mathematical models is subdivided into neural network, signal processing, pattern recognition, and expert systems.

 

View Alternative Text

Figure 2. Diagram. Structural parameter-identification system tree.

 

Parameter Identification with a Mathematical Model

If a mathematical model exists, or it is possible to make reasonable assumptions about structural connectivity and general properties, an FEM is useful. An optimization-based method can be used to adjust FEM parameters so that the differences between the analytical and experimental data are minimized. Major differences between the FEM and experimental results are classified as damage. At present, most of the static or dynamic FEM excitations are limited to producing linear-elastic responses of a structure regardless of any damage that structural components might have experienced.

Static Testing: Static testing is based on measured deformations induced by static loads. Comparatively little research has been performed using static loads, although static testing is analytically and computationally simpler than dynamic testing. Static testing can be easily applied on bridges using stationary truck loads or lateral pullback forces. Either displacements and rotations measurements or strains measurement can be determined. The displacements and rotations measurements can be used easily in FEM parameter estimation algorithms; however, static testing has limited general applications because most practical excitations are dynamic in nature.

Sanayei and Nelson proposed parameter estimations using static displacements and rotations measured at a subset of degrees of freedom (DOF).(51) Sanayei and Onipede expanded this work to estimate parameters with static loads applied at one subset of DOFs and displacements measured at another.(52) In both studies, structural element stiffness was successfully estimated including element failures in stable structures. Hajela and Soeiro successfully developed parameter estimations using simulated static deflections and vibration modes.(53) Banan and Hjelmstad(54) and Banan et al.(55,56) estimated stiffness parameters using incomplete sets of static load and displacement measurements. Bruno formulated a parameter identification technique to locate and characterize loose joints of a deployable space truss using actuator-induced static loading and unloading.(57)

Strains on a structural element are caused by both bending and axial deformations; therefore, strain measurements can capture structural element behavior, as well. Warren and Malvar monitored temperature strains and strains caused by slow-moving cranes on naval piers.(58) Sanayei and Saletnik developed parameter estimations for linear-elastic structures, using static strains and preserving structural connectivity.(59) Numerical simulations on truss and frame structures demonstrated the ability to identify all or a portion of structural cross-sectional properties including element failures.

Dynamic Testing: Most previous experiments have been done with dynamic excitations. Parameter estimations can be performed in either time or frequency domain to obtain structural parameters with identified or measured modal properties.

Directly using dynamic testing data in the time domain prevents further contamination resulting from data processing. Ibrahim and Mikulcik developed the algorithm that uses free vibration response of a structure to determine its modal parameters.(60) Ibrahim and Pappa applied this technique to large modal testing.(61) Seibold and Fritzen applied the Extended Kalman Filter technique to a nonlinear system and estimated unknown states and parameters simultaneously.(62) The Extended Kalman Filter technique might show poor convergence because of poor initial estimates or inaccurate assumptions regarding noise. A filter can be tuned to reduce the problems. Hjelmstad et al.(63) developed a robust time-domain estimator using velocities and displacements derived from the measured accelerations. This method can accommodate structural responses sampled incompletely in time, state, and space, and it is applicable to complex structural systems. Ghanem and Shinozuka evaluated four of the most popular methods for time domain parameter estimations with experimental data in a laboratory.(48) Based on the expertise required for each method and the quality of the estimated results, the method of recursive least squares with exponential memory was found to be the most promising.(64)

Dynamic testing data in the frequency domain is more compact than that in the time domain, and it more readily reveals structural modal properties. Earlier work by Baruch,(65) Berman and Nagy,(66) Chen et al.,(67) Collins et al.,(68) and Kabe(69) has laid the groundwork for more recent applications. Beck and Jennings estimated optimal modal parameters using a minimized outputerror function and earthquake ground motion.(70) Because of noise and limited test locations, only dominant modes were identified. Natke and Yao developed a damage-detection method based on multiple data sets where each set was obtained from different locations on a structure.(71) Stubbs and Osegueda developed a theory using changes in modal properties in beams, plates, and shells to detect damage.(72) The validity was demonstrated with cantilevered specimens with known damage.(73) This theory was refined and applied to offshore structures by Kim and Stubbs.(74) Dascotte was able to successfully update structural models using modal properties.(75) Smith and Beattie developed a method for optimal estimation of model parameters using inconsistent modal test data for large space structures.(76) Chen and Lurie used active members instead of external forces to induce vibrations and measured displacement responses.(77) Modal parameters were identified with equal or better accuracy than with conventional excitation tests. Fritzen and Zhu measured TFs by exciting mechanical models with broadband impulse spectrums, and the FEMs of the real structures were updated successfully.(78) Gornshteyn used selected frequencies and incomplete mode shape measurements for optimal parameter estimation at the structural element level.(79) Aktan et al.(3) successfully correlated the FEM and bridge load rating for a three-span continuous bridge using modal test data. Banan and Hjelmstad developed a method with sparsely sampled data in space, time, and state.(54) Liu minimized the norm of the modal force error to obtain the elemental properties of a truss structure.(80) Olson et al. performed modal testing on a bridge with unknown foundations.(4) Capabilities of modal testing to identify the unknown foundation characteristics were not confirmed. Farrar et al. studied the I-40 bridge over the Rio Grande River in Albuquerque, NM, with both FEM and experimental data.(81) The researchers concluded that the analytical and measured resonant frequencies or modal shapes were not sensitive to the damage through the plate girder.

Strains do not relate to mode shapes directly, but Yao et al.(82) developed a damage-detection method using vibration signature analysis and the concept of "strain mode shapes."(83) Strain measurements were found to be more sensitive to local damage and better at identifying damage than were displacement measurements.

Parameter Identification without a Mathematical Model

Parameter identification can be performed without a mathematical model in terms of the measured input and output, and the underlying mechanics can be treated as a black box. These methods are neural network, signal processing, pattern recognition, and expert system.

Ignoring the underlying mechanics involved with a problem, a neural network attempts to match a predefined pattern. Wu et al. performed a feasibility study to explore the use of neural networks in structural damage assessment.(84) Successful application requires that the neural network be trained to identify both damaged and undamaged structural behavior. Case studies at a threestory shear frame showed promising results. Masri et al. trained neural networks to predict the response of a damped Duffing oscillator.(85) The neural network successfully predicted the response of this nonlinear oscillator for both deterministic and stochastic excitations; however, significant issues must be resolved before neural networks can be applied to complex structures.

Signal processing produces structural modal properties from the experimental data; however, it is difficult, if not impossible, to identify the specific location of the damage without a mathematical model. Shinozuka et al. used the autoregressive moving average (ARMA) model to simulate multivariate random processes and prescribed correlation function matrices.(86) Unlike the Fast Fourier Transform (FFT) method, the analysis using the ARMA representation is not limited to computer memory availability. Signal processing can be used to extract the structural modal properties, which can then be used as input to a parameter-estimation system to identify structural element properties.

Kim and Stubbs used hypothesis testing and statistical pattern recognition for damage location.(74) They demonstrated the proposed algorithm to locate and estimate the severity of damage on a 41-member simplified jacket-type offshore platform. Stubbs et al. verified this algorithm on an instrumented multispan bridge located in New Mexico.(87) The damage was located accurately with three modes of vibration and no prior knowledge of the bridge material properties. This algorithm requires a mathematical model to identify the mode shapes; however, the damage location scheme is performed with pattern recognition and does not require a mathematical model.

An expert system can use both quantitative and qualitative data to assess the severity of damage in a structure. Because an expert system relies on expert judgment, it may not be possible to locate damage that is not apparent from a visual inspection. Ross et al. quantified the linguistic variables used to describe damage as fuzzy sets.(88) The fuzzy set, numerical data, and rules comprise the knowledge base for the expert system. The proposed bridge substructure evaluation and monitoring system will not use an expert system, so more examples are not provided.

Huang et al. recently developed the HHT, a technique for applying time domain data that makes it possible to analyze vibration data and determine the resonant frequencies of systems instantaneously by location throughout a time domain record for nonlinear, nonstationary systems.(5) In other words, the HHT method helps determine short-duration changes in the system response frequencies that indicate the lower frequency resonance associated with damage to a structural member. This technique promises to be more sensitive to short-term changes through lower frequency, nonlinear responses when a moving or varied excitation force is most actively exciting and closest to a damaged member. Thus, the masking of lower frequency responses associated with damage to a single member is better analyzed using the HHT approach. This comparatively new technique, applied late in this research, is discussed in chapter 7.

Uncertainty in Structural Parameter Identification

The purpose of structural parameter identification is to use measured data to determine the bestguess values of preselected unknown parameters. Uncertainty in structural parameter identification results from inherent errors and noise in the measured data, modeling errors, and some techniques that inherently estimate a biased parameter.

Regardless of the type of testing data used for parameter identification, sensors are subject to measurement noise. In the presence of noisy measurements (input error), the parameter identification algorithm will produce results different than those with noise-free simulated input data. The difference between the estimated parameters and true values (output error) is evaluated to determine the effect of input errors for a specific subset of measurements. It is economical to use as few inputs (applied forces and measured responses) as necessary, but each set of inputs has its own input-output error relationship; some sets of measurements are more sensitive to errors than others.(89,90) Therefore, designing a practical parameter identification system requires a careful selection of sensor locations to produce meaningful results at minimum cost. Because sensor location selection has a large effect on input-output errors, the most promising subset of measurements must be selected.

Haftka and Adelman proposed two integer program methods for selecting actuator locations to correct surface distortions of an orbiting spacecraft.(91) DeLorenzo proposed an improved version of the heuristic method, published in 1990.(92) DeLorenzo’s sensor and actuator configuration was used for control of large space structures. Hajela and Soeiro proposed the idea of dominant displacements for both static and dynamic testing, indicating that certain forces and responses are more representative of the structural system.(53) They also showed that errors are more prevalent when loading does not result in an equal stress distribution in each of the structural elements. In a related paper, they showed through the experimental results that a uniform stress loading produces excellent results.(93) Kammer developed the method of effective independence for sensor location selection for on-orbit modal identification.(94) A method developed by Holnicki- Szulc et al. was based on the progressive collapse analogy for optimal locations of actuators controlling the selected vibration modes.(95) Lim determined that damage is most easily detected for the elements that are fully participating in load bearing and contain the most energy.(96) Damage in such elements appears in the parameter estimation results. If there is a systemwide change in the estimated parameters, individual element damage cannot be isolated. Sanayei et al. studied the effect of measurement noise on the parameter estimates, and the researchers proposed a heuristic best-in-worst-out method for the preselection of static forces and displacement measurements based on error sensitivity analysis.(97) Sanayei and Saletnik applied the method to static strain measurements.(98)

Uncertainty in modeling and algorithms is inevitable, and further study in performing successful parameter identifications is needed. Brown developed a parameter identification procedure to estimate structural model parameters using measured responses and expert judgment.(99) Fuzzy updating was used to supply subjective information when numerical measurements were unavailable. Beck proposed a statistical system identification procedure that used averaged response measurements to estimate the structural model parameters.(100) Gangadharan et al. proposed a probabilistic system-identification method to infer structural model parameters of flexible joints.(101)

Damage Indices

A measure of damage or a damage index is required to make a rehabilitation decision. This information helps interpret the stiffness changes the parameter estimation procedures identified. A damage index, often normalized to a fixed scale for structural integrity, is an essential criterion in the damage assessment process.

Most damage indices available have been developed from research in the earthquake engineering community. During an earthquake, a structure is expected to experience inelastic response; consequently, it is possible to measure factors such as ductility and dissipated energy. Researchers Newmark and Rosenblueth,(102) Bertero and Bresler,(103) Banon and Veneziano,(104) Park and Ang,(105) Stephens and Yao,(106) Yao and Munse,(107) and Chung et al.(108) developed damage indices based on ductility ratios, cumulative damage, and component damage. Because of the redundancy of engineered structures, the damage estimate obtained for a simple structural element does not necessarily correspond to the damage sustained by the structural system. Bertero and Bresler,(103) Park and Ang,(105) DiPasquale et al.,(109) Pandey and Barai,(110) and other researchers have expanded their element damage indices to structural system damage indices with the use of weighting functions. Chung et al. discussed the importance of understanding the use of a structural damage index in its derivation.(108) To determine the integrity of a full structural system, the local damage indices for each element in the structural system can be combined with a Monte Carlo analysis to determine the probability of a structure’s failure. Nondestructive bridge condition assessment is performed either by visual inspection or with controlled or operating excitation, and it requires elastic structural responses. In a bridge management system, engineers commonly conduct visual inspections to assign a damage index to a bridge. Similarly, structural identification results must be interpreted on a comparable scale in a bridge management system. Aktan et al. related bridge damages to the incremental increase of structural flexibility.(3) Farrar and Cone determined that damage to a bridge superstructure must be significant before the global dynamic properties are affected.(111) Mayes applied the structural translation and rotation error-checking algorithm to locate the damage on a complex bridge that crosses the Rio Grande River in New Mexico.(112) Stubbs et al. located the damage in the same Rio Grande bridge by successfully using the first few dynamic modes, a baseline structure, and pattern recognition.(87)

Not much research has been done on bridge foundation identification and condition assessment. Reese and Stokoe performed a study on what instrumentation should be used for pile axial load tests.(113) Richart and Whitman conducted a study that compared experimental and theoretical results for a model footing foundation.(114) They found that footings on a semi-infinite elastic plane might be modeled as a mass-spring-dash pot system. Samtani et al. found that 2-D stress elastic-plastic finite element analysis underestimates the pier resistance to ship impact because it does not account for the wedging resistance in the plane strain analysis and the side friction on the out-of-plane sides of the caissons.(115) They presented a methodology that helps extrapolate 2- D models to account for 3-D effects. Further research is needed in this area.

In addition to the efforts outlined above, the interest in instrumentation to monitor the health and condition of structures is not limited to the civil infrastructure research community. In fact, research has been underway in areas such as aerospace, mechanical, electrical, computer, and systems engineering. Certain elements of this research form reasonably mature fields on their own, such as Norman et al.,(116) Pau,(117) Patton et al.,(118) and Willsky.(119) Therefore, it is logical and potentially advantageous to approach instrumented bridge health and condition monitoring from an interdisciplinary point of view.

A damage index provides information about the state of a structure when the index was measured. Long-term structural deterioration is defined as the change in value of structural parameters at incremental stages during the design life of a structure. Such changes in structural parameters can be represented as time-dependent damage indices. Damage accumulates and reaches a certain level such that the structure becomes deficient. Structural deterioration models can be used to predict the change in structural parameters or damage indices considering the intended structural loads, environmental conditions, maintenance practices, and historical data.

Bridge Management Systems

In the past 2 decades, the relationships between bridge conditions and rehabilitation decisions have begun to be formalized in the form of BMSs. A BMS is an integrated collection of the following elements:

BMSs can be used to accomplish the following tasks:

BMSs can be used to answer the following example questions:

As new types of data are generated and the relationships between structural condition and vulnerability to natural hazards are better understood, BMSs can assist in preservation and improvement decisions during the lives of bridges.

The current managerial focus for a BMS requires the ability to plan and forecast maintenance needs or rehabilitation procedures. The two most widely known BMSs, Pontis®(120) and BRIDGIT©, are described in Czepiel(121) and Egri et al.(2) The history of bridge management is also described in Czepiel.(121) The next three paragraphs briefly review changes and advances in three areas: the aging bridge inventory, regulations, and technical advances. Examples illustrate the need for BMSs and their capabilities.

Aging Bridge Inventory

Although the condition of the Nation’s bridges gradually improved over the period from 1992 to 2004, the percent of deck area on deficient bridges still exceeds the targets, which in 2004 were 26.4 percent for national highway system bridges and 28.8 percent for non-national highway system bridges.(122) Meanwhile, highway travel grows annually, and financial resources are highly constrained at all levels of government. The 50th anniversary of the interstate highway system will be observed in 2006. The system is aging, and bridge owners have expressed concern about problems caused by age, fatigue, impact, and the environment. These issues have encouraged many States to pursue the development of BMSs as a systematic method for managing some of the problems.

Regulatory Constraints

The Intermodal Surface Transportation Efficiency Act (ISTEA) of 1991 required States to develop and use BMSs.(123) The requirements were detailed in an interim final rule effective January 3, 1993.(124) The rule was issued as interim "because of concerns about the data burden that states, metropolitan planning organizations (MPO), and local agencies may have."(125) Based on comments and reports from many agencies of burdensome experiences, ISTEA, as of November 1995, did not mandated BMSs; however, FHWA encourages agencies to implement the systems and required States to report on progress by October 1996.(123) The spirit of ISTEA is evident as numerous States continue to pursue BMS development.

Technical Advances

Several technical advances are likely to have significant effect on state-of-the-art BMSs. These advances range from faster, larger, more portable, and more flexible computer systems, to new data collection technologies and sensors for data acquisition. In addition, advances in software, graphical user interfaces, and multimedia technologies facilitate integration of BMSs into organizations.

The Role of Bridge Substructure Condition in Bridge Management Systems

In a BMS, the role of bridge substructure condition varies with the specific implementation. In the National Bridge Inventory, bridge substructures have an overall rating based on visual inspection. Each State should maintain, at a minimum, a history of bridge substructure condition rated visually.

Beyond the National Bridge Inventory rating, Pontis provides bridge substructure as a default element category definition; specific elements can constitute a condition. Elements can be user defined or drawn from a list of default element definitions that include piers, pier walls, and abutments of various materials (concrete, reinforced concrete, masonry, and timber). Condition states and actions for each element are also provided. For example, a reinforced concrete submerged pile (element 227) in condition state 3 (exposed steel) would require cleaning and patching (action item 41). Pontis also allows users to define condition units as parts of the same element that are in different environments and may have different conditions.(126)

BRIDGIT is similar to Pontis in inventory and condition rating. Additional user-defined data items can be added to the inventory and condition information is included as a percentage of a particular element in a given condition state. BRIDGIT appears to offer some additional flexibility in bridge-specific condition data such as load rating. The limits of this flexibility warrant further exploration.

In general, visual inspection is the widely used means for including condition data in a BMS.(121) This has several disadvantages. First, it is qualitative, and ratings generally do not exhibit a high degree of consistency or repeatability. Second, the rating reflects an aggregate measure of condition. For example, microscopic flaws can be catastrophic, but they will never be reflected in the rating. Third, the ratings are not closely related to the cause of the problem or the response. Finally, the rating accuracy is unknown.

While there is clearly a role for improving and enhancing the status of bridges in this country through better NDE and its integration with BMSs, several issues must be addressed.

A dynamic bridge substructure evaluation and monitoring system is just one aspect of nondestructive testing appropriate for bridges. How valuable is this information compared with other NDE methods? Answering this question requires consideration of the cost of NDE.(127)

Because BMSs currently include only visual condition data, the value of multiple sources of data needs to be explored. NDE data might not substitute for visual condition ratings, but could complement, reinforce, or support visual condition rating data.

Much of the work on dynamic bridge evaluation is in search of a specific type of defect. No systematic attempts have been undertaken to analyze dynamic testing data in terms of the effect on bridge life cycle costs, the opportunities to reduce the probability of such defects in similar bridges, and the effectiveness of remedial strategies.

The influence of low probability, high impact catastrophic events on bridge life cycle costs need to be considered. Bridge engineers have been successful at avoiding catastrophic failure but the cost must be accounted for. (128) The burdens of data collection and analysis also should be explored.(128)

The opportunities to improve BMSs using a dynamic bridge substructure evaluation and monitoring system lie in the following areas:

Recent work on structural reliability, data needs for a BMS, and expert systems such as RETAIN, an expert system for retaining wall rehabilitation, may provide some insight into opportunities to integrate into BMSs in use or under development a dynamic bridge substructure evaluation and monitoring system.(129, 130, 131)

 

Previous | Table of Contents | Next

Federal Highway Administration | 1200 New Jersey Avenue, SE | Washington, DC 20590 | 202-366-4000
Turner-Fairbank Highway Research Center | 6300 Georgetown Pike | McLean, VA | 22101