Distress Data Consolidation Final Report

As discussed in chapter 2, eight distresses are included in the consolidated data set for each surface type. With eight distresses for each of the 7,428 surveys, 59,424 individual data elements were reviewed. Of these 59,424 data elements, less than 10 percent (5,398) were discrepant; of the 5,398 discrepant data elements, more than half appear to be easily resolvable.

All discrepancies were classified into one of three categories as shown in table 13. Each discrepant data set was reevaluated to try to identify commonalities within the three categories. The DIM interpretation and summarization explained in chapter 2 were classified as "Variability Associated with Human Error." The discrepancies that fell into the categories of seasonal effects or data collection methodology were believed to have several different causes, as listed in table 13.

A recommendation was made to FHWA that the discrepant surveys should be reevaluated to correct any errors that may exist. This chapter includes a description of each category and suggestions as to how the discrepancies might be resolved.

Table 13. Categories and probable causes of discrepant survey data.

VARIABILITY ASSOCIATED WITH HUMAN ERRORS

Misunderstanding of Distress Definitions

Misunderstanding of distress definitions was the most commonly occurring and obvious source of discrepant data, causing 23 percent of the discrepancies. Due to the subjective nature of visual condition surveys, it is understandable and logical that such discrepancies would occur. The following sections of this chapter describe some of the scenarios identified under this heading and offer recommendations for resolution.

Cracking in the Wheel Path Versus Fatigue Cracking

While one individual may record distress as fatigue, the next individual may record it as longitudinal cracking in the wheel path. An example of this discrepancy is shown in appendix A (figure 5). The DIM defines fatigue as cracking that usually occurs in the wheel path and has a measurable width.⁽²⁾ In some cases, the presence of a measurable width is debated: one surveyor may indicate that the cracking does have width, while another surveyor indicates that the cracking does not have width.

Of the questionable non-wheel path longitudinal cracking data, 31 percent were identified as questionable fatigue cracking, 20 percent were identified as questionable wheel path longitudinal cracking, and 1 percent were identified as questionable non-wheel path longitudinal cracking. It was not anticipated that the non-wheel path longitudinal cracking would be affected by this discrepancy. However, cases were found where a combination of errors occurred, including an error affecting the differentiation between wheel path and non-wheel path longitudinal cracking.

Several potential solutions exist for resolving this discrepancy, the simplest of which is majority rules. If three of four surveys classify the cracking as fatigue, then the fourth should be amended to record it as fatigue also. Conversely, if three of the sections show longitudinal cracking in the wheel path, the fourth should be amended to indicate longitudinal cracking in the wheel path. Another solution would be to delete the outlier from the data set. A third potential solution would be to combine the longitudinal cracking in the wheel path and fatigue cracking for all existing surveys, as has been previously recommended. Any of these recommendations could be implemented without revisiting the survey maps.

Cracking in the Wheel Path Versus Non-Wheel Path

There are sections where longitudinal cracking varies from wheel path to non-wheel path and back again. In other words, one surveyor will decide that the cracking occurs in the wheel path, while the next surveyor will record that the cracking occurs outside the wheel paths. In these instances, re-evaluation of the maps is required to resolve this discrepancy.

Appendix A (figure 6) illustrates a section for which the longitudinal cracking appears to wander from wheel path to non-wheel path with time. In more than 25 percent of the HMA surveys, the distinction between non-wheel path and wheel path longitudinal cracking was a problem.

Block Cracking Versus Longitudinal and Transverse Cracking

Several data sets were identified in which the distinction between block cracking and extensive quantities of transverse and longitudinal cracking was the source of the problem. Generally, block cracking is expected to follow a regular pattern, which is not necessarily true of extensive transverse and longitudinal cracking. Still, the percentage of questionable surveys in which block cracking was an issue was significantly lower than those in which longitudinal or transverse cracking was an issue: 2 percent for the former, versus more than 15 percent for the latter.

This discrepancy is illustrated in appendix A (figure 7). Difficulty in distinguishing between block cracking and longitudinal and transverse cracking accounted for approximately 53 percent of the questionable block cracking data.

Unfortunately, solutions to this discrepancy are limited because it is difficult to separate block cracking into longitudinal and transverse cracks, or to combine longitudinal and transverse cracks into block cracking. However, if the majority of the data points reflect block cracking, it is conceivable that a re-review of the distress maps could produce a quantity of block cracking from the recorded longitudinal and transverse cracking for entry into the consolidated distress data set.

Errors in Summarization

A summarization error is an error in determining the total quantity of distress from the map created during the distress survey. Summarization errors are estimated to be the source of questionable surveys as follows: 1 percent of the questionable transverse cracking number, 19 percent of the questionable transverse cracking length, 4 percent of the questionable patching number, and 5 percent of the questionable patching area. Because errors of this type are fairly obvious from a review of the distress maps, they are easily corrected.

For some of the surveys, it is evident that a summarization error was made. Examples include obvious mismatches between the numbers of cracks or patches recorded and the total area (e.g., 20 patches for a total area of 0.2 square meters (m²) is unlikely). In some instances, the quantities exceed a practical limit (i.e., patching greater than 15 percent of the surface area prompted removal of the section from the study). The assumption is that a math error occurred in totaling the distress to provide these mismatches between two related data elements, so that distress magnitudes this great should be a rare occurrence.

Appendix A (figures 8 and 9) displays two cases where a summarization error appears to have occurred. In the first case, shown in figure 8a, a survey showed that almost 30 transverse cracks were apparent on the test section, but the cracks had 0 length as shown in figure 8b. In the second case, a section was shown to have approximately 15 patches (as shown in figure 9a) with a total area of about 500 m² (as shown in figure 9b). Recall that FHWA requirements call for placing a test section out-of-study if more than 15 percent of the area is patched, so that distress magnitudes this great should be a rare occurrence.

VARIABILITY DUE TO SEASONAL EFFECTS

None of the seasonal effects shown in table 13 were evident from this preliminary review. To confirm that seasonal effects were the source of the questionable distress data, the distress maps must be reevaluated. In using these maps, it might be possible to determine that a particular crack found during one survey was not found in another. Furthermore, recording the general site conditions at the time of the survey might help to explain oversights or errors in the distress noted. However, without a more thorough review, it is impossible to state categorically that the cause of the discrepancy noted is due to a specific seasonal effect. Although seasonal effects may be the source of some discrepancies, such could not be confirmed in this exercise.

VARIABILITY ASSOCIATED WITH COLLECTION METHODOLOGY

While none of the anticipated discrepancies noted in table 13 were apparent for the key distresses examined in this work, there was at least one discrepancy that could be attributed solely to methodology.

In the early rounds of PASCO film review and manual surveys, no distinction was made between wheel path and non-wheel path longitudinal cracking. When the DIM was revised in 1993, the distress maps resulting from manual surveys performed prior to that date were reviewed, and the longitudinal cracking was classified as wheel path or non-wheel path. This distinction has not yet been made for data collected under the PADIAS 1.x system. For the purposes of this review, all longitudinal cracking for PADIAS 1.x was considered as non-wheel path longitudinal cracking. This assumption is not valid for all of the PADIAS 1.x data.

Appendix A provides an example of this discrepancy (figure 10). This discrepancy affected 159 (16 percent) instances of the questionable non-wheel path longitudinal cracking and 189 instances (21 percent) of the questionable wheel path longitudinal cracking.

In reevaluating the questionable surveys of longitudinal cracking (wheel path and non-wheel path) from the PADIAS 1.x method, the data sets have been denoted as requiring "Clarification of Longitudinal Cracking Position." To resolve this particular discrepancy, the PADIAS 1.x film for these sections must be reevaluated.

It was anticipated that there would be a consistent bias in the data collection methodologies (e.g., manual surveys consistently yielded greater distress than automated surveys). Although in some surveys all distress quantities are high or all distress quantities are unusually low, few were outside the tolerable error limits set. None of these extremes could be attributed exclusively to a particular methodology.

MISCELLANEOUS SOURCES OF DISCREPANT SURVEY DATA

There are additional causes for discrepant survey data that did not fit the aforementioned categories. These include insufficient quantities of distress, nonlinear growth rate, and documentation of maintenance and rehabilitation activities. Fortunately, these are the easiest to identify and rectify.

Insufficient Quantities of Distress

For some sections, surface distress was virtually nonexistent, making it impractical to compute realistic measurements of variability and error. Moreover, the error band has a practical lower limit. Below this limit, it is not possible for the distress to be measured within the error bands and still allow for some realistic measurement error. While the error band process used in the graphical review accommodated many of these concerns, there were still several for which all of the points were zero except one, and for the one non-zero point the error bands did not encompass zero.

Figures 11, 12, and 13 in appendix A illustrate this concept for the HMA, JC, and CRC surface types, respectively. Table 14 includes the number and percentage of surveys for which there were insufficient quantities of distress to identify any performance trends.

It is recommended that these data sets be reclassified as acceptable, citing insufficient quantities of distress.

Nonlinear Growth Rates

In evaluating the data, it was assumed that distress would increase in a generally linear fashion with time. Those data sets that were initially labeled as discrepant due to inordinately high quantities of distress on later surveys were reclassified as acceptable to allow for nonlinear increases in distress.

An example of nonlinear growth rate is illustrated in figure 14 in appendix A. The discrepancy was observed only on the HMA-surfaced test sections. Table 15 includes the number and percentage of data in which a nonlinear increase in distress was recorded.

Undocumented Maintenance or Rehabilitation

The plots generated in the graphical review process reflected any maintenance or rehabilitation activities that had occurred on the test sections that were recorded in the database. For some of the test sections, all of the distress dropped to near zero from the previous survey, with no rehabilitation or maintenance noted. Rehabilitation and/or maintenance are the most likely cause(s) of this drop when observed on several distresses.

A data analysis feedback report listing these sections was supplied to FHWA for review and resolution of these discrepancies. Figure 15 in appendix A illustrates this source of error.

Table 14. Number of elements affected by insufficient quantities of distress

Table 15. Number of elements affected by exponential growth rates.

Summary

Table 16 includes a listing of the discrepancies and probable cause of each. As stated earlier, less than 10 percent of the 59,424 data elements reviewed were discrepant. Of the 5,398 discrepant data elements, more than half may be addressed easily with only two exceptions: 1) distinction between block cracking and combined longitudinal and transverse cracking, and 2) those discrepancies that did not readily fit the categories established in this review. From this preliminary review, the authors are confident that, with minimal additional effort to correct these discrepancies, at least 95 percent of the critical distress data could be suitable for incorporation into the consolidated data set.

Table 16. Number of discrepant surveys for each category.

HMA

JC

CRC

Guideline Key

1. Insufficient quantities of distress.
2. Nonlinear growth rates.
3. Combination of fatigue and longitudinal cracking, WP.
4. Undocumented maintenance or rehabilitation.
5. Re-review of PASCO 1.0 longitudinal cracking.
6. Discrepancy between block cracking and longitudinal and transverse cracking.
7. Error in summarization.
8. Wander in longitudinal cracking from WP to NWP.

99. Other (or discrepancy was unable to be categorized).