Featuring developments in Federal highway policies, programs, and research and technology.
|This magazine is an archived publication and may contain dated technical, contact, and link information.|
|Federal Highway Administration > Publications > Public Roads > Vol. 71 · No. 6 > History Lessons From the National Bridge Inventory|
Publication Number: FHWA-HRT-08-004
History Lessons From the National Bridge Inventory
by Waseem Dekelbab, Adel Al-Wazeer, and Bobby Harris
Analyzing data from the NBI can help predict how bridge decks will perform.
The Federal Highway Administration (FHWA) is charged with working with States to keep the Nation’s highway system operating safely, including the approximately 590,000 bridges that cross rivers and roadways and otherwise link the system together. After the 1967 collapse of the Silver Bridge over the Ohio River between West Virginia and Ohio, which resulted in 46 deaths, the U.S. Congress and the President established a law that marked the beginning of uniform standards for bridge inspections. The law includes mandates for States to inspect all public road bridges of a certain length and requires the U.S. Secretary of Transportation, in consultation with the States, to maintain an inventory of these bridges: the National Bridge Inventory (NBI).
August 1, 2007, offered another terrible reminder that the Nation cannot take bridge safety for granted. During the evening rush hour, the main spans of the I–35W Mississippi River bridge in Minneapolis, MN, collapsed, killing 13 people and injuring nearly 100 more. Construction of the bridge concluded in 1967, the same year the Silver Bridge collapsed, and 40 years later it had become Minnesota’s fifth busiest, carrying 140,000 vehicles daily.
Researchers recently analyzed the numbers and data stored in the NBI. Their findings could offer insight and improve understanding of bridge performance based on 24 years of information compiled in the database. For example, information gained from this research will help answer the question, “How much longer is a bridge in a certain condition likely to stay in that condition before deteriorating further?”
Twenty-Four Years Of Information
State departments of transportation (DOTs) manage their infrastructure assets through pavement and bridge management systems that monitor conditions cyclically, measure performance, and predict future trends and requirements based on deterioration models and other analyses. Assessing bridge performance, however, still relies heavily on expert opinions and significant assumptions and generalizations.
Like other long-term investments, bridges must be maintained for safe and reliable public service. Aging, traffic demands, environmental impacts, and budget limitations jeopardize the safety of bridges, which makes incorporation of best practices in asset management imperative.
“The current lack of understanding of bridge performance and the unavailability of high-quality data on the various facets of bridge performance are the two most important factors that severely limit the application of useful management tools, such as life-cycle cost analysis, and hinder the ability to direct resources to the most critical issues in bridge performance,” says John M. Hooks, bridge technology consultant, Highway R&D Services.
Researchers recently used information from the NBI to investigate bridge conditions over time and then collected the results into one data set. The goal was to improve understanding of bridge performance empirically, using a time-based, probabilistic approach (that is, looking at the probability that a condition will change over a certain period). By following an empirical approach, the researchers sought to reduce the number of assumptions in deterioration modeling. They investigated selected bridge characteristics — such as deck construction method (cast in place versus precast) and type of wearing surface (concrete or asphalt overlay) — and traffic variables that contribute to bridge deterioration, with an eye toward leveraging the data in the NBI to enhance models for predicting deterioration.
The National Bridge Inspection Standards (NBIS), in place since the early 1970s, require biennial safety inspections for highway bridges longer than 6.1 meters (20.0 feet) on public roads. State and local DOTs collect and report information on the conditions and compositions of the structures. Baseline information includes descriptions of the functional characteristics, location, geometry, ownership, and maintenance responsibilities for each bridge.
This information enables FHWA to characterize the highway bridge system on a national level and to analyze bridge composition and condition. Safety is ensured through periodic hands-on inspections and ratings of the primary components of a bridge, such as the deck, superstructure, and substructure. The data in the NBI drive the funds apportioned for the highway bridge program.
FHWA maintains the composition and condition information stored in the NBI. This database is the most comprehensive source of information on bridges in the United States, with some 14.3 million inspection entries over the 24-year period from 1983 to 2006.
Deterioration Models For Bridge Decks
Predicting performance is the main challenge in the life-cycle assessment and asset management of bridges. Performance of the bridge deck is a major maintenance and serviceability concern because the deck is the component most prone to problems that affect traffic and requires the most maintenance and replacement work. Loss of deck performance generally results from corrosion (caused by natural salinity or direct application of deicing agents), traffic loading and vibration, temperature fluctuations, and other factors.
Deterioration models that describe deck performance are based on either a deterministic or a probabilistic approach. A deterministic approach essentially deals with known (certain) outcomes, such as adding two or more numbers and arriving at a definite sum. Most models, however, use the probabilistic approach, which deals with outcomes that are not exactly known (uncertain) and uses statistical terms to describe them, such as taking the average of several figures, because of uncertainty about deterioration behaviors and pertinent causes. The probabilistic approach can be divided into state-based and time-based models. State-based models predict the probability that a bridge deck will have a change in its condition during a fixed interval, and accumulate this probability over multiple intervals, as in a Markov chain model. Time-based models predict the probability distribution of the time taken by a deck to change its condition to the next lower condition.
State DOTs have used Markovian (state-based) models for bridge deterioration since the early 1990s, following guidance in the American Association of State Highway and Transportation Officials’ (AASHTO) AASHTO Guide for Commonly Recognized (CoRe) Structural Elements. Many bridge management systems, such as Pontis and BRIDGIT use Markovian models to predict the condition of bridge elements, such as bridge deck joints and reinforced concrete beams, based on expert judgment.
Markovian deterioration models are “memoryless,” meaning the predicted future condition of a bridge component depends only on its current condition and not on previous behavior. This is an advantage since the model does not require a full time series of inspections over the life cycles of a population of bridges. On the other hand, this approach is disadvantageous because it does not include the time elapsed in the initial condition.
Tapping the NBI
According to the researchers, NBI data might show improvement in bridge conditions that could be the result of actions such as maintenance, rehabilitation, or reconstruction. The only action reported in NBI data, however, is the reconstruction year. These actions, described by the researchers as observed improvement, could lead to improvement in the observed bridge conditions from one inspection year to the next. Statistical analysis of the NBI data, based on the aforementioned definition of observed improvement, shows that the average age of a bridge before observing the first condition improvement is approximately 7 years.
The researchers used data from 1983 through 2006 in this analysis. Given that the first data year in the analysis is 1983, the researchers operated under the assumption that bridges built between 1976 and 1982 showed no prior condition improvement, without observed improvement, during their first deterioration cycle, which continued up to or beyond 1983. On the other hand, the researchers assumed that bridges built before 1976 would have condition improvements and related deterioration cycles occurring despite observed improvement.
Number of Deck Conditions Versus Observed
The total number of bridges with a complete set of historical inspection records is 210,204. Each bridge history has many deterioration cycles based on observed improvement status. For the 210,204 bridges, there are 705,068 condition transitions (including decks both with and without observed improvement). By dividing the total number of condition transitions with observed improvement (641,710) by the number of bridges with complete inspection records (210,204), the researchers determined that there is an average of about three condition transitions per bridge with observed improvement.Source: FHWA.
The researchers using the NBI bridge data studied changes in the bridge conditions in the time-series data to investigate the effects of observed condition improvement on bridge deterioration behavior even though the database has no information on the type of improvement action. The researchers did not exclude from the database and analysis any records that show improvement or deterioration in bridge deck condition.
In the computed time-series database, the researchers generated variables that provide the following information for each bridge: time in each condition, time in each deterioration cycle, number of deterioration cycles, and bridge condition before and after an observed improvement. The engineers used relational database linkages to associate each bridge and component condition with deterioration cycle identifiers and the conditions before and after the observed improvement for a given deterioration cycle. With these variables and linkages, the researchers were able to generate a new summary analytical time-series database.
Another important step in data preprocessing is data censoring. The researchers denoted a condition that has an unknown start date, reflects a bridge built before 1983, or extends through 2006 (the time of the end of this research) as a censored condition. Their censoring process replaced the censored condition’s time in condition with the national historical average time in condition. The researchers applied censoring to all bridge deterioration cycles spanning these two data boundaries to enable comprehensive use of all data available in the NBI database.
The researchers integrated the computed time-in-condition database for bridge decks with selected bridge characteristics from the NBI. Based on the integrated data set, researchers can examine whether those characteristics affect deck performance in terms of time in condition.
The researchers independently investigated a limited number of bridge characteristics, including deck structure type (NBI Item 107), type of wearing surface (NBI Item 108A), average daily traffic, or ADT (NBI Item 29), maintenance responsibility (NBI Item 21), and year built (NBI Item 27). The time-series data analysis is a time-based model using the Kaplan-Meier method, which often is applied to medical statistics to determine patient survival probabilities, to find the survival function, or curve. The medical community uses survival curves to estimate a patient’s prognosis by comparing it to the experiences of other patients in like situations who received similar treatment. For highway bridge decks, a survival function represents the probability that a bridge deck remains in its condition state for at least time (t). The survival function curves show the percentage of bridges maintaining a given condition versus the time in this condition.
Each bridge in the NBI historical time-series database has gone through one or more deterioration cycles that consist of one or more condition rating changes. Different bridges in a given condition have different times in the condition, leading to a unique probability density function of time in condition. From the probability density function, the researchers calculated the cumulative distribution function of time in condition. The survival function of a given condition is equal to 1.0 minus the cumulative distribution function.
The last step in this approach was to compare the survival curves for each variation within each bridge characteristic (for example, differ— ent wearing surface types) and for different bridge conditions in order to investigate bridge performance empirically. A number of advanced numerical approaches could be applied, such as the expected value method, linear regression, or Poisson regression, but the researchers intentionally avoided these approaches to reduce the number of assumptions needed and to focus on the empirical approach.
Analysis: Deck Survival Curves and Findings
Before considering any particular bridge deck characteristic in the data analysis, the more general case of survival functions (condition change) for two deck groups — those with observed condition improvement and those without — is telling. Deck condition and observed improvement status clearly affect bridge survival function or time in condition.
For decks with conditions 9, 8, and 7 — that is, they are in relatively good condition — without observed improvement, the historical trend shows the probability to remain in these conditions is higher than for decks with observed improvement. The data show the opposite is true for decks in condition 6 and lower — those in fair or poor shape — that is, they fare better with observed improvement.
A dramatic drop in deck survival probability for bridges without observed improvement occurs in conditions 9 and 8 after 3 and 4 years, respectively. The best explanation is that deterioration begins but is difficult to observe initially in a visual inspection, but once manifested can appear more widespread or advanced, and thus more easily observed.
Further, the survival curves of decks in different conditions with observed improvement have a much narrower band of survival variation (similar survival performance) than decks without observed improvement.
Deck Structure Type
The percentage of condition transitions grouped according to deck type is 74 percent for concrete cast-in-place decks, 12 percent for wood decks, 7 percent for concrete precast panel decks, 3 percent for corrugated steel decks, and 4 percent for other types.
The survival curve for the major deck types without observed improvement, ranked from highest to lowest in terms of survival probability, is as follows: precast concrete panels, cast-in-place concrete, corrugated steel, and wood. In the case of observed improvement, the researchers found nearly the same trend except that the wood deck has a higher survival probability than corrugated steel, meaning that wood survives longer in a given condition than corrugated steel.
As with deck groups as a whole, deck structures in condition 7 without observed improvement have a higher survival probability than deck structures with observed improvement. This trend does not hold, however, for bridge deck structures in condition 6 and lower, where decks with observed improvement have a higher survival probability than those without, especially for time in condition less than 10 years. That means that bridges in condition 6 and lower, with observed improvement, have a greater survival probability.
Deck Wearing Surface Type
The percentage of condition transitions grouped according to wearing surface type is 36 percent for decks with bituminous (asphalt) surfaces, 31 percent for monolithic concrete (cast at the same time as the bridge deck by adding extra concrete thickness), and 33 percent for other types, including (low-slump concrete, wood, and gravel). Analysis of survival curves shows that monolithic concrete decks in condition 9 have higher survival probability than decks with bituminous wearing surfaces in condition 9, with or without observed improvement. Bridge decks in condition 6 or less with monolithic or bituminous wearing surfaces show a higher survival probability for the group with observed improvement than the group without it. In other words, some wearing surface types are more effective in protecting bridge decks than others.
Average Daily Traffic
Traffic is a variable widely understood to affect bridge deck performance. The researchers developed survival functions according to the following ADT classes: less than 100 vehicles; 100–1,000 vehicles; 1,000–10,000 vehicles; 10,000–100,000 vehicles; and more than 100,000 vehicles. Analysis supports the logical expectation that survival probability for decks in condition 9 is higher when the ADT class is smaller, regardless of observed improvement status. With fewer vehicles crossing it, a bridge deck is likely to last longer.
For ADT greater than 100,000, the number of deck condition changes observed in the data is relatively small compared with the other ADT classes. The data contained only 600 deck condition transitions — less than 1 percent of the total data. This result likely is a reflection of the fact that a relatively small proportion of bridges across the Nation are exposed to more than 100,000 vehicles per day, and thus fewer observed deck conditions and condition transitions are available for study.
Type of Deck Structure and Wearing Surfaces
Bridge Year Built
The researchers grouped the survival functions according to each bridge’s construction date, within 10-year intervals starting from 1900 and ending with the truncated 2001–2006 period. Some 67 percent of deck condition changes affected bridges built between 1950 and 1990. The researchers found only 1.5 percent of transitions in decks built after 2000. This finding is simply a reflection of the aging nature of the Nation’s bridge infrastructure. That is, bridges built recently have short histories, having been in service just 5 to 6 years; moreover, the number of bridges constructed recently is small in comparison to the number built between 1950 and 2000.
Analysis of the survival curve reveals that old bridge decks show higher survival probability than young bridge decks, but this is a result of young bridges’ limited history. This result might be due to defining bridge age based on year built instead of the time that the most recent rehabilitation or replacement took place.
In the case of bridge decks with observed improvement, the newer bridges show a higher survival probability in the first 3 years (for reasons already explained), and after that no clear trend is observed for those bridges, although each bridge’s survival curves become closer to each other as the bridges age. Bridge construction date as a survival factor requires more investigation: Data quality might be a factor if the bridge deck was replaced without recording the reconstruction date in the NBI.
Examining the NBI clearly shows the importance of time-in-condition history for predicting the probability of condition change. The researchers found that many factors influence time in condition, such as ADT, where lower ADT corresponds to increased bridge deck time-in-condition survival functions. Bridge decks with concrete precast panels show higher time-in-condition survival functions than other deck types, and the researchers found a similar trend for decks with monolithic concrete wearing surfaces.
Bridge decks in good shape (conditions 8 and 9), or apparently good shape, remain that way without observed improvement for the first 3 or 4 years, then the survival rate drops dramatically because initial deterioration processes likely are not observed by visual inspection in the early stages.
Taking Advantage of NBI Information
The NBI, which contains more than 14 million inspection entries over a 24-year period from 1983 to 2006, offers valuable information for bridge management research. Bridge owners can use this information to better understand deterioration trends based on empirical data and to minimize the uncertainty in predicting bridge conditions and, as a result, enhance decisionmaking.
If bridge owners are to use the NBI more efficiently, then FHWA, State DOTs, and other stakeholders need to address the data quality issues, such as gaps in the data record, out-of-range values, or input codes that do not conform to the NBI coding guide. One issue, condition rating fluctuations (cycles of decrease in condition rating followed by 1 or more years of increase in condition rating), might be related in some cases to data quality instead of resulting from maintenance. In this preliminary study, the researchers assumed that condition rating fluctuations are the result of some unknown maintenance action or activity. To create the time-in-condition time-series database, they included only those bridges with complete condition rating histories over the 24-year period.
This preliminary investigation demonstrates that further research could glean more bridge performance data from the extensive historical inspection record contained in the NBI. Future research could extract additional knowledge by performing a multidimensional analysis to isolate influences on bridge deterioration. External factors such as environment and natural hazards also could be integrated with the NBI time-series database to enhance bridge performance knowledge. Moreover, other advanced statistical analyses and data-mining tools could help researchers take advantage of the NBI’s empirical data set.
Waseem Dekelbab, Ph.D., PE., is a senior bridge research engineer at bd Systems, Inc., a subsidiary of SAIC, and serves as principle investigator in the Bridge Management Information Systems (BMIS) Laboratory at FHWA’s Turner-Fairbank Highway Research Center (TFHRC).
Adel Al-Wazeer, Ph.D., is a senior research engineer with bd Systems, Inc., who has been working in the BMIS Laboratory at TFHRC since 2000. Al-Wazeer holds a doctorate degree in civil engineering from the University of Maryland, College Park. His doctoral dissertation title is “Risk-Based Bridge Maintenance Strategies.” He received the Eisenhower Grant for Research Fellowship from the National Highway Institute and the International Road Federation Fellowship.
Bobby Harris is a senior project manager at bd Systems, Inc., and serves as the contractor’s project manager for the BMIS Laboratory at TFHRC. He holds a B.S. in geology and M.S. in geophysics from North Carolina State University and has worked in transportation research and transportation information systems for more than 20 years.
For more information, contact Waseem Dekelbab at 202–493–3451 or firstname.lastname@example.org, Adel Al-Wazeer at 202–493–3202 or email@example.com, or Bobby Harris at 703–999–4807 or firstname.lastname@example.org.
For access to the data, contact the BMIS Laboratory at www.tfhrc.gov/bmis/contacts.htm.
This article represents the views of the authors and not the position or policies of FHWA.
Page Owner: Office of Corporate Research, Technology, and Innovation Management
Scheduled Update: Archive - No Update
Technical Issues: TFHRC.WebMaster@dot.gov