Reliability of Visual Bridge Inspection
Visual inspection techniques are the primary methods used to evaluate the condition of the majority of the nation's highway bridges. These subjective assessments may have a significant impact on the safety and maintenance of a bridge. However, until a study was performed at the Federal Highway Administration's Nondestructive Evaluation Validation Center (NDEVC), a complete study of the reliability of these inspections had not been undertaken.
This article is the second of two on the visual inspection study conducted at NDEVC and describes the results of this recently completed study. The first article, "Studying the Reliability of Bridge Inspection," appeared in the November/December 2000 issue of Public Roads (Vol. 64, No. 3) and describes how the study was conducted.
The general approach taken in the field investigation was to have a representative group of practicing bridge inspectors complete a battery of predefined inspection tasks at NDEVC test bridges. The subject population consisted of 49 bridge inspectors from 25 state departments of transportation. These inspectors were asked to complete seven routine inspections and three in-depth inspections at NDEVC test bridges while being monitored by NDEVC staff. Information about the inspector and the inspection environment was collected to assess their influence on inspection reliability.
Study Results
Routine Bridge Inspection Results
One set of information generated during a routine inspection is a series of "condition ratings" assigned to the bridge deck, superstructure, and substructure. These condition ratings give an overall measure of the condition of a bridge by considering the severity of deterioration in the bridge and the extent to which it is distributed throughout each component. The ratings assigned to each element are based on a standard set of definitions associated with numerical ratings between zero (failed) and nine (excellent condition). Inspection agencies can use these ratings to track deterioration and to allocate maintenance funds.
Table 1 - Routine Inspection Condition Rating Statistical Information | ||||||||
Bridge | Element | Average | Standard
Deviation |
Mini-
mum |
Maxi-
mum |
Mode | N | Reference Rating |
B521 | Superstructure
Substructure† Deck |
5.9
6.1 5.8 |
0.78
0.79 0.81 |
4
3 3 |
8
7 7 |
6
6 6 |
49
49 49 |
5
6 5 |
B101A | Superstructure
Substructure Deck |
4.2
4.3 4.9 |
0.77
0.76 0.94 |
2
3 2 |
6
6 7 |
4
4 5 |
49
49 48 |
4
4 4 |
B111A | Superstructure
Substructure Deck |
4.6
5.5 5.2 |
0.86
0.77 0.92 |
2
4 3 |
7
7 7 |
5
5,6 6 |
49
48 49 |
4
5 4 |
B543 | Superstructure
Substructure Deck |
5.3
6.1 4.8 |
0.88
0.89 0.94 |
4
4 2 |
7
8 6 |
5
6 5 |
44
47 48 |
5
6 5 |
B544 | Superstructure
Substructure Deck |
5.8
5.3 4.5 |
0.72
0.83 0.74 |
4
3 3 |
7
7 6 |
6
5 5 |
48
47 48 |
6
6 4 |
Route 1 | Superstructure
Substructure Deck |
6.7
7.2 7.1 |
0.66
0.57 0.53 |
5
6 6 |
8
8 8 |
7
7 7 |
49
49 49 |
7
8 7 |
Van Buren‡ | Superstructure†
Substructure Deck |
6.8
6.7 5.8 |
0.64
0.62 0.92 |
6
6 4 |
9
8 7 |
7
7 5 |
24
24 24 |
7
8 7 |
† The condition rating results for this element did not pass C2 test for goodness-of-fit.
‡ This task was performed by a team of two inspectors who could collaborate to reach their findings. |
The condition rating results for the seven routine inspection tasks are given in table 1. This table summarizes the average, standard deviation, maximum, and minimum condition rating results from the participating inspectors. A reference rating is also given in the table. This is the condition rating given to each element following thorough assessments by NDEVC inspectors. Further analysis of these data revealed that the condition ratings were normally distributed in all but two instances.
Note that table 1 includes the inspection results generated during the inspection of the Van Buren Road bridge. For this inspection, inspectors were allowed to work in two-person teams and to collaborate in determining their inspection results. The following discussion will exclude these results and will focus on condition ratings assigned by individual inspectors, except where otherwise noted.
Table 2 - Results of the t-Test at Five-Percent Significance Level for the Average Condition Ratings | ||||||
Bridge
|
||||||
Element |
B521
|
B101A
|
B111A
|
B543
|
B544
|
Route 1
|
Deck |
Fail
|
Fail
|
Fail
|
Pass
|
Fail
|
Pass
|
Superstructure |
Fail
|
Fail
|
Fail
|
Fail
|
Fail
|
Fail
|
Substructure |
Pass
|
Fail
|
Fail
|
Pass
|
Fail
|
Fail
|
Pass = average inspector Condition Rating and reference Condition Rating can not be considered statistically different.
Fail = average inspector Condition Rating and reference Condition Rating are statistically different. |
Because the reference ratings given in table 1 and the inspector-assigned ratings were frequently different, a statistical analysis was performed. This analysis examined whether or not the two ratings were statistically different by applying what is known as the t-test. In a t-test, "fail" indicates that the two ratings are different, and "pass" indicates that the two ratings are the same from a statistical standpoint. From table 2, it is apparent that in most cases, the average inspector condition ratings are statistically different from the reference ratings.
The distribution of sample condition
ratings was found to be normal; an example of which is shown in figure 1, indicating that the sample condition ratings can be used to predict how the general population of bridge inspectors would perform. From these analyses, it was found that 95 percent of condition ratings would be assigned over a distribution of five discrete ratings or ±2 from the mean. Furthermore, only 68 percent would be distributed over three discrete ratings or ±1 from the mean.
During the inspection of the Van Buren Road bridge, the inspectors were asked to use their respective state inspection forms, and several teams also submitted element-level inspection data. Element-level inspections rely on specific definitions of elements to classify the bridge structure and to describe any observed deterioration using the defined condition states. One of the most common element-level inspection systems uses the Pontis bridge management system, but other systems also exist. Fourteen inspection teams reported results consistent with the commonly recognized (CoRe) elements. The CoRe element system is a standardized set of descriptions of common bridge elements and conditions.
The major deck, superstructure, and substructure elements were used very consistently by each of the teams reporting element-level data. The "other superstructure/substructure" elements were recorded much less consistently. As an example, there was significant confusion in the use of CoRe elements for the joints with three different definitions being used to describe the same joint. Another example is bridge railings, which were defined by three different elements, with only three out of 14 teams defining the element correctly.
As expected, the greatest variability in the element-level inspection data occurred with the non-CoRe elements. For example, five teams used five different elements to track wingwall information. Another four teams used five different elements to track slope protection.
Although not necessarily required, inspection notes were often generated during an inspection to supplement condition ratings and/or condition state assignments. As such, the use of inspection notes was investigated. Although the inspectors participating in this study may have taken a large number of inspection notes during each of the tasks, this analysis focused only on a small set of notes deemed to be of principal importance. These notes generally described poor to very poor condition elements. Although not described here in great detail, when analyzed, it was found that most inspectors made note of the severe deficiencies, but typically, at least one-fifth of the inspectors did not note a specific condition. It should be pointed out that the level of deterioration precipitating each of these notes is so severe that one could expect a nearly 100-percent note-taking rate.
One other way to document inspection findings is through the use of photographs. During one of the inspections, the inspectors were provided with a camera and asked to document their findings. The photographs could generally be grouped into 18 different types. Of these 18 photographs, 13 were identified as needed to fully document the bridge conditions. On average, each inspector took just over seven photographs with a maximum of 19 and a minimum of one. The wide variability in the number and specific types of photographs taken illustrates the differences in the documentation policies of the agencies as to what constitutes "full" documentation for a routine inspection.
One important aspect of the experimental study was the quantitative measurement of human and environmental factors thought to potentially have a relationship with inspection results. The quantitative measurements were made through a series of written questionnaires, oral interviews, environmental measurements, and first-hand observations. The relationship of these factors to the inspection results was then studied. A multivariate, nonlinear analysis was required to find correlation between these factors and the inspection results. The analysis revealed that several factors appear to have a relationship with the inspection results. Specifically, the inspector's fear of traffic; near visual acuity; color vision attributes; formal bridge inspection training; and the inspector's perception of the bridge's maintenance, accessibility, and complexity were found to have a consistent relationship with inspection results.
In-Depth Inspection of Steel Superstructure Bridges
The in-depth inspection tasks of the superstructures of two steel bridges performed during this study were intended to provide insight into the accuracy and reliability of close-up, hands-on inspection performed by bridge inspectors. As the goal of this type of inspection is the specific identification of global and localized deficiencies, the accuracy and reliability were studied in the context of correctly noting the presence of known deficiencies.
During the in-depth inspection of bridge B544, inspectors were asked to inspect approximately one-fifth of the superstructure of this riveted plate girder bridge. To access the bridge, inspectors were provided with a ladder and a man-lift. Of the 49 inspectors participating in the study, 42 completed this inspection.
Two basic classes of deficiencies are present in bridge B544. First, there are general, recurring deficiencies: paint system failure, moderate to severe corrosion, and extensive corrosion and section loss of rivet heads. Second, local deficiencies also exist: an implanted crack indication at the root of a tack weld, a missing rivet head, impact damage at two locations, and an abnormal rocker bearing rotation.
Table 3 summarizes the defect-detection results for the notable deficiencies. These data show that the inspectors reported the general, recurring deficiencies with a relatively high frequency, however, a much lower percentage of inspectors noted the local deficiencies. For example, all inspectors reported the paint system failure, which is obvious throughout the structure. However, only half of the inspectors noted bearing misalignment, and only three inspectors noted a crack indication.
The second in-depth inspection of a steel superstructure bridge was performed at the Route 1 bridge. The inspectors were asked to inspect a single bay of one span of the bridge. As with bridge B544, the inspectors were allowed to use a man-lift to gain access to the superstructure during the inspection. The Route 1 bridge is a medium-span bridge with 1.83-meter-deep welded plate girders. The superstructure framing consists of welded transverse and longitudinal stiffeners, bolted angle diaphragms, bolted and welded flange transitions, and a lateral bracing system of angle and T-members bolted to lateral gusset plates welded to the girder web.
The Route 1 bridge has deficiencies in three general categories: general deficiencies, welded connection deficiencies, and bolted connection deficiencies. Obviously, the welded and bolted connection deficiencies pertain to those specific connection types, and the general deficiencies include all other deficiencies. Specifically, the general deficiencies are paint system failure, corrosion, member distortions, and fabrication errors. The welded connection deficiencies consist of crack indications that occur in or close to a weld. In the Route 1 bridge, there were four recurring locations that were likely to produce welded connection deficiencies. These locations had seven crack indications. Three bolted connection deficiencies occurred at cross-frame-to-vertical stiffener connections in the form of bolts with nuts at least four millimeters removed from the plate that they were to bear against.
Table 3 also summarizes the deficiency-detection results for the Route 1 bridge. In total, the overall accuracy rate for correctly identifying crack indications was only 3.9 percent. In addition, there was a false-positive rate of 0.6 percent for identifying non-cracked welds as having crack indications. With respect to the bolted connection deficiencies, the overall accuracy rate was 24 percent with a false-positive rate of 0.5 percent. As with the results from the in-depth inspection of bridge B544, these data indicate that a far greater percentage of inspectors identify the general deficiencies than the local deficiencies. In fact, deficiencies such as crack indications were correctly identified by less than 20 percent of the inspectors. More than half of the inspectors noted more general conditions such as corrosion and paint failure.
While each inspector was performing the inspection of the Route 1 bridge, NDEVC staff noted how the task was performed and what specific items were inspected. This information was used to make a pseudo-quantitative measure of the thoroughness of each inspector with respect to the inspection of welded connections. To accomplish this, four parts of the test bridge were considered based on the locations that were likely places for crack indications to occur. Inspectors were assigned rating points contingent on the thoroughness of the inspection of these areas. This rating system allows each inspector to achieve a rating between zero and 10 based on the overall thoroughness of the weld inspection.
The inspector thoroughness ratings were used to classify the inspectors into profile groups. The groups were defined as those inspectors who received a score of eight to 10, those who received a score of five to seven, and those who received a score of zero to four. Forty-five percent of the inspectors earned a rating of eight or higher. These inspectors could be considered to have completed a thorough in-depth inspection of the superstructure. Of the inspectors who correctly identified a crack indication, 86 percent were from this group. Eighteen percent of the inspectors earned a rating of between five and seven, and thus, they are considered to have completed an in-depth inspection on part of the structure. Fourteen percent of the inspectors who correctly identified a crack indication were from this group. Finally, 36 percent of the inspectors earned a rating between zero and four; these inspectors can be considered to have performed an incomplete in-depth inspection. None of the inspectors in this group correctly identified a crack indication.
Table 4 shows the results by profile groups for a number of factors. These factors are summarized because they are correlated with the inspectors in the three inspector profile categories. Various trends in the table are evident. Specifically, the inspectors who earned the higher profile ratings tended to take longer to complete the inspection, were generally more mentally focused, and were more comfortable than average when performing the inspection. These inspectors were also more likely to use a flashlight, to expect fatigue-related deficiencies, and to be closer to the welds that they were inspecting. The converse is true for each of these factors for the inspectors who earned the lower inspection profile ratings.
Bridge Deck Delamination Assessments Using Mechanical Sounding
A delamination assessment was conducted on the two southern spans of the Van Buren Road bridge. This assessment was conducted by teams of two inspectors using only visual inspection techniques, including mechanical sounding. The Van Buren Road bridge has a 175-millimeter-thick concrete deck that has significant delaminations with very few visible indications that those deficiencies exist.
Of the 22 teams of inspectors completing the assessment, 20 provided maps of the delaminations that they found, and some of the teams also provided a numerical estimate of the amount of delaminated area. The two remaining teams provided only a numerical estimate. In total, five teams were within five percentage points of the delamination percentage determined by NDEVC - 19 percent of the deck area. Fourteen teams were within 10 percentage points, and all of the teams were within 20 percentage points. Three teams indicated that the deck was less than 5 percent delaminated. Although these teams fall within 20 percentage points of the correct delamination percentage, it is obvious that these teams failed to detect large areas of the deck that were delaminated.
The relationship between areas of the bridge deck that the inspectors indicated to be delaminated and the areas found to be delaminated by NDEVC is also indicative of the accuracy of this type of deck inspection. Figure 2 provides the delamination results for the 20 teams that produced delamination maps. This figure presents the locations where various numbers of teams indicated the presence of a delamination.
As can be seen in this figure, most teams performed relatively poorly at locating individual delaminations. In total, 69 percent of the deck was indicated to be delaminated by at least one inspection team. In addition, only approximately 1 percent of the deck was indicated as delaminated by at least 15 teams, and no areas were indicated as delaminated by all inspection teams. By examining the shading levels, the consensus of five or more teams shows the best correlation with the delamination area found by NDEVC. The consensus of five or more teams showed 21 percent of the deck delaminated compared to 19 percent determined by NDEVC.
Concluding Remarks
Although significant advances have been made in the development of nondestructive evaluation technologies, visual inspection is still the predominant tool used to assess bridge conditions. However, this study shows that there are aspects of bridge inspection that need significant improvement. For routine inspections, condition ratings, element-level inspection results, inspection notes, and photographs are used with significant variability. Of greatest importance is the amount of variability found in the assignment of condition ratings. For in-depth inspections, it appears that when an in-depth inspection is prescribed, the inspection may not yield any findings beyond those that could be noted during a routine inspection. The results of the deck-delamination survey conducted during this investigation indicate that this type of inspection does not consistently provide accurate results.
Brent M. Phares, Ph.D., is a research engineer for Wiss, Janney, Elstner Associates Inc., a consultant to the Infrastructure and Inspection Management Team in the Office of Infrastructure Research and Development at the Federal Highway Administration's Turner-Fairbank Highway Research Center in McLean, Va. He received his doctorate in structural engineering from Iowa State University.
Dennis D. Rolander is a principal research engineer for Wiss, Janney, Elstner Associates Inc. He received his master's degree in structural engineering from North Carolina State University.
Benjamin A. Graybeal is a research engineer for Wiss, Janney, Elstner Associates Inc. He received his master's degree in structural engineering from Lehigh University.
Glenn A. Washer, Ph.D., is the program manager of the Federal Highway Administration's Nondestructive Evaluation Validation Center at the Turner-Fairbank Highway Research Center in McLean, Va. He has a master's degree in civil engineering from the University of Maryland and a doctorate from The Johns Hopkins University. Washer is a licensed professional engineer in Virginia.
The authors thank the inspectors who participated in the field portion of the visual inspection study; the input of the inspectors was invaluable. In addition, the authors gratefully acknowledge the contributions to the study made by Alabama, Alaska, Arizona, California, Colorado, Delaware, District of Columbia, Florida, Georgia, Hawaii, Illinois, Indiana, Kansas, Kentucky, Louisiana, Maine, Maryland, Massachusetts, Michigan, Minnesota, Mississippi, Missouri, New Hampshire, New Jersey, New Mexico, New York, North Carolina, North Dakota, Ohio, Oklahoma, Oregon, Pennsylvania, Rhode Island, South Carolina, South Dakota, Tennessee, Texas, Utah, Vermont, Virginia, Washington, Wisconsin, and Wyoming.