Pavement Performance Measures and Forecasting and The Effects of Maintenance and Rehabilitation Strategy on Treatment Effectiveness (Revised)

CHAPTER 4. DATA MINING AND SYNTHESIS

DATA SOURCES

The data used in this study were obtained from the LTPP database Standard Data Release 28.0. The database contains six volumes consisting of the primary dataset, data compilation views, FWD measurements, profile data, traffic data, and LTPP Traffic Analysis Software tables. Each of the six volumes contains various data elements for the more than 2,500 pavement test sections included in the LTPP Program. At the time of this study, about 1,700 test sections had been de-assigned or decommissioned from the LTPP Program over time, but nearly 800 remained active under the various experiments. Planning and scheduling has been taking place under the direction of FHWA to establish additional experiments and test sections to study different/new topics. Table 17 and table 18 list the number of active test sections under the SPS and GPS experiments, respectively, at the time of this report. The data from both active and de-assigned test sections were extracted from the database and arranged in special format for analyses. The detailed data extraction is presented in later sections in this chapter. Results of the analyses are presented and discussed in chapters 5, 6, and 7.

Table 17. Active LTPP SPS test sections as of January 2014.

Table 18. Active LTPP GPS test sections as of January 2014.

In addition, the pavement management databases from three State transportation departments—CDOT, WSDOT, and LADOTD—were requested and received. From each database, several pavement projects were identified, and their data were downloaded and formatted for analyses. Each of the selected projects was subjected to certain treatments in the past. The data for each project included the location reference systems, the time-series pavement conditions and distresses, the time and types of treatments that were performed in the past, and, in some cases, the cost of the treatments. The data were analyzed and the results are discussed in chapter 9.

AUTOMATED AND MANUAL PAVEMENT DISTRESS DATA

The monitoring module within the primary dataset of the LTPP database contained time-series pavement distress data (rut depth, cracking, and so forth). The data were collected using two different survey procedures: manual (visual observations) and semi-automated (videotape).

Table 19 and table 20 list the number of manual and semi-automated surveys conducted for each test section in the SPS-1 experiment. The data for all other test sections in the SPS and all test sections in the GPS experiments were submitted to FHWA and are available from the LTPP Customer Support Services.⁽⁷⁹⁾ After detailed examination of the manual and the semi-automated pavement distress and condition data, the manual data were selected for data modeling and analyses. The semi-automated data were not used for the following reasons:

• The number of available manual data points was much greater than for the semi-automated data. The manual data had been collected over the entire duration of the LTPP Program, while the semi-automated data were only collected between 1989 and 2004. Hence, less semi-automated data were collected.
  
  
• The two sets of data were not compatible enough to be combined and analyzed as a function of time. The few semi-automated data points generally did not align with the trends indicated by the manual data over time.
  
  
• The variability of the time-series semi-automated pavement distress and condition data was much greater than that of the manual data.

It is important to note that similar findings were reported by another team of researchers.⁽⁷⁶⁾

Table 19. Number of manual and semi-automated surveys for test sections 0101–0112 in LTPP SPS-1 experiment.

Table 20. Number of manual and semi-automated surveys for test sections 0113–0124 in LTPP SPS-1 experiment.

DATA EXTRACTION

To facilitate the analyses of this study, the following specific data items were extracted from the LTPP database for each test section of the SPS and GPS experiments and formatted for time-series analyses:

• Inventory.
  
  
• Time-series pavement condition and distress.
  
  
• Time and type of pavement rehabilitation, preservation, and maintenance actions.
  
  
• Traffic.
  
  
• FWD data, including those collected along the test sections in the SMP.
  
  
• Climatic region.

Inventory Data

All inventory data, including construction history of the test sections, their opening dates to traffic, lane widths, number of lanes, pavement layer types and thicknesses, and subgrade information, were obtained from the inventory module. Some of the tables specifically used for this purpose were INV_AGE, INV_GENERAL, INV_ID, INV_LAYER, INV_SUBGRADE.

Time-Series Pavement Condition and Distress Data

The time-series pavement condition and distress data used in the analyses (presented in chapter5) included transverse, longitudinal, and alligator cracking; rut depth; IRI; and faulting. These data were extracted from their respective files and reorganized in a spreadsheet format for analyses. The pavement condition and distress data were obtained from the following LTPP tables under the monitoring module of the LTPP database:

• Cracking (MON_DIS_AC_REV, MON_DIS_CRCP_REV, and MON_DIS_JPCC_ REV): The cracking data were classified and stored in the database using three severity levels—low, medium, and high—as described by the LTPP Distress Identification Manual.⁽⁸⁰⁾ The difficulty with such data was that the crack severity rating was a function of several variables, including the following:
  
  
  • The pavement temperature at the time of the distress survey. The crack width, which is part of the severity level assignment, is a function of the pavement temperature. In general, the crack width increases as the temperature decreases.
    
    
  • The subjective judgment of the surveyor who reviewed and observed the cracks. Such subjective judgment is a function of the degree of training and experience of the surveyors. Further, the same pavement segment is likely to be surveyed by different surveyors over time. Thus, a crack may be labeled high severity in one year and medium the next year or vice versa. Figure 43 and figure 44 depict, respectively, the time-series low, medium, and high severity transverse cracking data of SPS-3 test section A330 and SPS-5 test section 0502 in California. Examination of the figures indicated that the sum of the time-series low, medium, and high severity cracking data was more consistent over time than the individual severity levels of cracking and therefore more suitable for modeling. Further, various attempts were made to analyze the data per severity level. In each attempt, the data for a significant number of test sections were eliminated from the analyses because of their high variability over time. In addition, the transverse cracking data were classified and stored in the LTPP database as either unsealed or sealed cracking at each severity level, while the longitudinal cracking data were classified and stored at each severity level as sealed and unsealed cracks in the wheelpath or non-wheelpath. The sealed and unsealed cracks had the same effect on the pavement structural integrity; the only difference was that sealed cracks retarded water infiltration, which might slow the rate of deterioration. Further, the wheelpath and nonwheelpath longitudinal cracks differed in their potential causes. Wheelpath longitudinal cracks in flexible pavement were likely to be either the start of top-down cracking due to pavement-tire interaction or the first appearance of alligator cracks on the pavement surface. In fact, for some test sections, the extent of longitudinal cracking, from one survey cycle to the next decreased substantially as alligator cracking was recorded for the first time, indicating that the longitudinal cracks were reclassified as alligator cracks. Therefore, for flexible pavements, longitudinal cracking in the wheelpath were combined with alligator cracking to facilitate the analyses of the data. Note that the selection of treatment type was based on the severity and location of the cracks, but the condition rating of the pavement was not affected by such information.
    
    

1 ft = 0.305 m.

Figure 43. Graph. Transverse cracking versus elapsed time for LTTP SPS-3 test section A330 in California.

1 ft = 0.305 m.

Figure 44. Graph. Transverse cracking versus elapsed time for LTTP SPS-5 test section 0502 in California.

• Roughness (MON_PROFILE_MASTER): The time-series pavement roughness data were computed into IRIs and stored in the database as left wheelpath IRI and right wheelpath IRI. The average of the two values was considered equivalent to the effect of roughness on the traveling vehicle. Hence, the average IRI was considered in the analyses.
  
  
• Faulting (MON_DIS_JPCC_FAULT): For transverse joints, pavement faulting data were stored as edge faulting and wheelpath faulting. The wheelpath faulting data were used in the analyses because faulting at the edges could be influenced more by warping and/or curling. In addition, the average faulting among all joints within a given test section was calculated and used in the analyses.
  
  
• Rut Depth (MON_T_PROF_INDEX_SECTION): In the initial stages of the LTPP Program, rut depth measurements were made using a 4-ft (1.2-m) straightedge reference under the assumption that wheel-path depressions were not wider than 4 ft (1.2 m). These rut depth measurements can be found in the MON_RUT_DEPTH_ POINT table. However, in many instances, the wheel-path depressions were wider than 4 ft (1.2) m. Hence, transverse profile measurements were chosen by the LTPP Program over straightedge measurements to account for this.⁽⁵⁹⁾ The transverse profile data were used by the LTPP Program to calculate the mean and the maximum rut depth in each wheelpath. The average of the two means and the average of the two maximum rut depth data were calculated, and the former was used in the analyses in this study.

Pavement Rehabilitation, Preservation, and Maintenance Data

Pavement rehabilitation, preservation, and maintenance data were extracted from the MAINT_REHAB module of the LTPP database. The treatments performed on the LTPP test sections were classified and stored under different tables, namely MNT_IMP and RHB_IMP. After downloading the data, the research team organized them such that the treatment information for each LTPP test section could be easily retrieved and analyzed. For all LTPP test sections that received one or more treatments and for each treatment type, the pavement condition and distress data were organized in two different groups—before treatment and after treatment. Such grouping was crucial to accurately model the pavement performance before and after treatment and to estimate the treatment benefits.

Traffic Data

The research team extracted the traffic data for each test section from the traffic module of the LTPP database. The equivalent single-axle load (ESAL) data in the TRF_ESAL_COMPUTED table were used to group the various test sections. The groupings were used to assess the impact of various variables on pavement performance, the longevity of the pavement sections, and the effectiveness of the pavement treatments.

FWD Data

The research team extracted FWD data from the FWD measurements folder of the LTPP database to analyze the possible relationships between the measured deflection and pavement distress and/or conditions. The peak pavement deflection measured at each of the seven or nine sensors for different loads (different drop heights) were extracted and organized for analyses. Further, the time-series LTEs were also extracted. Finally, the FWD data of all test sections included in the SMP were extracted and organized for analyses. Because, for each test site included in the SMP, the FWD data were collected at different time and temperature, the data were analyzed to do the following:

• Check the accuracy of existing temperature correction procedures and develop a global temperature correction function.
  
  
• Determine whether the time-dependent FWD data could be used to establish a threshold value to trigger structural treatment or as an indicator of the structural integrity of the pavement sections.

Climatic Data

North America has four climatic regions: DF, DNF, WF, and WNF. The climatic regions were obtained from the TRF_ESALS_INPUTS_SUMMARY table of the traffic module in the LTPP database. The criterion established by the LTPP Program to identify wet and dry climates was based on annual precipitation. Regions with annual precipitation of less than 20 inches (50.8 cm) per year were considered dry. The classification of freeze or no freeze was based on the Freezing Index. Test sites in regions where the annual Freezing Index was greater than 150 degree-days were considered in a freezing climatic region.

STATUS OF THE CONDITION AND DISTRESS DATA

As stated in the previous section, for each LTPP test section in the SPS and GPS experiments, and for each pavement treatment type, all available before treatment and after treatment condition and distress data were downloaded and organized in spreadsheet format for analyses. Table 21 summarizes the alligator, longitudinal, and transverse cracking data of all SPS-1 test sections located in Montana (State code 30). Each row in the table represents one pavement test section. The columns indicate the treatment types and the number of pavement condition and distress surveys (i.e., number of time-series data points) that have been conducted before and after each treatment. For example, test section 0113 was subjected to crack sealing, aggregate seal coat, and two additional crack sealing treatments since its assignment to the LTPP Program. The number 5 under the first before treatment column indicates that five time-series data points (surveys) were available in the LTPP database taken before the first crack sealing treatment was applied. The number 2 under the after treatment/before treatment column indicates that two data points were available in the database taken after the first crack sealing treatment and before the aggregate sealing treatment. The numbers under the other after treatment/before treatment columns indicate the number of data points available in the LTPP database that were taken after the previous treatment and before the next treatment. Finally, four time-series data points were available in the LTPP database taken after the last crack sealing treatment. Note that the number of before treatment data points taken before the first crack sealing application and the number of after treatment data points taken after the last crack sealing treatment were greater than three, and hence, the data can be modeled as a function of time.

A similar summary for each SPS and GPS experiment in each State was made. The summaries and the information listed in table 21 were used to identify the test sections and the treatments for which three or more time-series condition or distress data points were available that were collected before and/or after a particular treatment.

Table 21. Number of cracking data points available before treatment and after treatment for LTTP SPS-1 test sections in Montana.

STATUS OF THE MAINTENANCE AND REHABILITATION DATA

As stated earlier, the maintenance and rehabilitation actions and their time of application were compiled for analysis. The number of test sections in the LTTP SPS and GPS experiments with and without treatments are summarized in table 22 and table 23.

Table 22. Number of LTTP SPS test sections with available treatment data in the database.

Table 23. Number of LTTP GPS test sections with available treatment data in the database.

Test sections in the LTTP SPS-1 through -7 and LTTP GPS-6, -7, and -9 were included in the analyses to assess the impacts of design variables and treatment benefits. The available data for the untreated test sections were used as control sections or to estimate the service period of the test section.

ANALYSES PROCEDURES

Most of the proposed analyses for the evaluation of the effectiveness of the various pavement treatments are based on the determination and evaluation of the relationships between the before and after treatment pavement performance. Such pavement performance is a function of the available time-dependent pavement condition and distresses data and the corresponding rates of pavement deterioration. To model, with some degree of certainty, the condition and distress data over time using nonlinear mathematical functions, a minimum of three time-series data points are required before and/or after treatment. Two or fewer data points do not define the parameters of the nonlinear mathematical functions representing the data. Examination of the available data points in the LTPP database indicates that, for a significant number of test sections, only two data points were available before and/or after treatment. To enhance the number of available data and to increase the number of test sections that could be analyzed, several actions were proposed in the interim report, discussed during the project meeting in Washington, DC, and implemented in the analyses. The actions taken are described in the following subsections.

Addition of One Data Point Immediately After Certain Treatments

Often, the pavement conditions and distresses were not measured immediately after construction or after treatment application. Depending on the treatment type, the condition and distress values after some treatment actions can be logically and reasonably assumed. Therefore, for all newly constructed SPS-1 and -2 test sections and for all other test sections where AC overlay or mill-and-fill treatments were applied, one can reasonably assume that at 0.01 years (3 days) after construction, the initial value of the rut depth, faulting, and the total length of each crack type would be negligible. Because a 0.0 data point is not allowed in the mathematical functions reported in the literature and used in modeling the data, the initial pavement distress and condition at the elapsed time of 0.01 years after construction were assigned the following insignificant values:

• Rut depth: 0.01 mm for the LTPP data and 0.01 inches for the State data.
  
  
• Transverse cracks: 0.01 m for the LTPP data and 0.01 ft for the State data.
  
  
• Longitudinal cracks: 0.01 m for the LTPP data and 0.01 ft for the State data.
  
  
• Alligator cracks: 0.01 m² for the LTPP data and 0.01 ft² for the State data.

This assumption supports the addition of one extra data point for use in the analyses of pavement performance. Unfortunately, no initial IRI can be reasonably assumed. To illustrate, in Oklahoma, skin patching was applied to 12 SPS-1 test sections. The LTPP database contained more than three time-series pavement condition and distress data points that were collected after the skin patching was performed, but only two data points were available that were collected before the treatment. Because all SPS-1 test sections were newly constructed, one data point could be assumed indicating that at 0.01 years after construction, the magnitudes of rut depth, crack length, and faulting were the same as those listed in the previous bullets. The addition of such data points made the analyses of the before treatment pavement performance possible. Once again, such an assumption was reasonable and logical because for flexible pavements, the smooth-drum rollers typically used in the compaction of the original HMA or overlays or mill-and-fill treatments produce smooth and flat pavement surfaces with no rutting or cracking. The LTPP treatments that were considered for this action are listed in table 24. This addition of a data point immediately after treatment was applied only to pavement segments where only two before treatment and/or after treatment data points were available. If fewer than two data points were available, the procedure would not yield three data points and hence it was not used. The addition of such data points significantly enhanced the number of available pavement segments for analyses. Note that no data points were added to any pavement segment that was subjected to any treatments not listed in table 24.

Table 24. Condition or distress type eligible for data addition for different treatments.

Table 25 summarizes the status of the cracking data of all SPS-1 test sections located in the DNF region. Similar tables for all pavement condition and distress data types and other LTPP experiments and climatic regions were submitted to FHWA and are available from the LTPP Customer Support Services.⁽⁷⁹⁾

Table 25. Summary of cracking data for LTPP SPS-1 test sections located in the DNF region.

The following describes the data that appear in each of the columns in table 25:

• Column A: Climatic region.
  
  
• Column B: State and State code.
  
  
• Column C: Treatment type.
  
  
• Column D: Number of SPS-1 test sections.
  
  
• Column E: Number of times treatments were applied. When the number in column E is greater than the number in column D, it implies that at least one test section received the treatment more than one time.
  
  
• Column F: Number of treatment applications for which three or more time-series data points are available that were collected before and after treatment.
  
  
• Column G: Number of treatment applications for which three or more time-series data points are available that were collected before treatment only.
  
  
• Column H: Number of treatment applications where three or more time-series data points are available that were collected after treatment only.
  
  
• Column I: Number of treatment applications where one data point can be logically assumed to have been collected immediately after treatment (as stated earlier as 0.01year), which yields three time-series data points before and after treatment.
  
  
• Column J: Number of treatment applications where one data point can be logically assumed to have been collected immediately after treatment (i.e., the addition of the 0.1‑year point described earlier in this section), which yields three time-series data points that were collected before treatment only.
  
  
• Column K: Number of treatment applications where one data point can be logically assumed to have been collected immediately after treatment (i.e., the addition of the 0.1‑year point described earlier in this section), which yields three time-series data points that were collected after treatment only.
  
  
• Column L: Number of test sections that can be analyzed before and after treatment.
  
  
• Column M: Number of test sections that can be analyzed before treatment only.
  
  
• Column N: Number of test sections that can be analyzed after treatment only.

At the time of this report, there were 1,555 LTPP test sections (supplemental sections were not included) in the SPS-1 through -7 and in the GPS-6, -7, and -9 experiments. The majority of these test sections (1,301) were treated at least 1 time during their assignment period. The total number of treatment applications was 2,674 (some test sections received more than 1 treatment). For new construction (SPS-1 and -2 test sections) and for overlay and mill-and-fill treatments, one rut depth, cracking length, and faulting data point was added at 0.01 years after construction.

Table 26 lists the number of test sections that could be analyzed before and after treatment, the number of test sections that could be analyzed before treatment only, and the number of test sections that could be analyzed after treatment after the addition of this data point.

Table 26. Summary of treatments applied to LTTP SPS-1 through -7 and GPS-6, -7, and -9 sections analyzed in this study.

Using the Control Section Data for Before Treatment Conditions

Several of the LTPP experiments, including SPS-3 through -6, were designed with a control section (untreated) adjacent to the test sections (which were subjected to various treatment types). The control section was subjected to almost the same traffic and environmental loading and had almost identical structure and subgrade support characteristics. For this reason, the performance data of each control section could be used to represent the before treatment performance data of the adjacent test sections when only two or fewer before treatment data points are available in the database for the given test section. For example, SPS-3 test section A310 (see figure 45) in Maryland had only one cracking data point (not shown in the figure) collected before an overlay treatment was performed. Six cracking data points were available after the overlay was performed. To analyze the before treatment conditions of test section A310, the performance data of the control section A340 (see figure 46), which was not subjected to treatment, was used to represent test section A310 before-treatment performance.

1 ft = 0.305 m.

Figure 45. Graph. Total longitudinal cracking versus time for LTTP SPS-3 test section A310 in Maryland.

1 ft = 0.305 m.

Figure 46. Graph. Total longitudinal cracking versus time for LTPP SPS-3 control section A340 in Maryland.

In some cases, where no control sections were assigned, the linked GPS test sections associated with the SPS sections were used as control sections.⁽⁶⁰⁾ Linked GPS test sections were under the GPS experiment and were located adjacent to the SPS test sections. They had traffic loading and structure similar to the SPS test sections to which they were linked. Some of the linked GPS sections were also treated. However, the before treatment data (see figure 47) could still be used as the before treatment data for the SPS-3 test section.

1 ft = 0.305 m.

Figure 47. Graph. Total longitudinal cracking versus elapsed time for LTPP GPS-1634 linked test section to SPS-3 experiment in Maryland.

Cracking data collected before the overlay could also be used as before treatment data for test section A310. Comparisons of the pavement condition and distress data between the control and the linked section that were related to the conditions of the test section were made to verify whether the data of the control and/or linked sections were indeed similar to the available before treatment data points of the test section. Finally, if the data of the control section represented the before treatment data of the test section, the performance of the two sections were compared to determine the benefits of the treatment applied to the test section.

For some test sections, such as LTTP SPS-3 test section A350 in New York, the reported before treatment longitudinal cracking was about 984 ft (300 m), as shown in figure 48. The longitudinal cracking data of the associated control section A340 indicated 197 to 328 ft (60 to 100 m) of cracking, as shown by the open symbols in figure 48. It should be noted that the control section was not subjected to any treatment. Nevertheless, in this and similar cases, the data from the control sections were not used because they were not representative of the before treatment pavement performance of the test section in question. Note that the use of the control section data in place of the before treatment data significantly increased the number of available test sections for analysis.

1 ft = 0.305 m.

Figure 48. Graph. Total longitudinal cracking versus elapsed time for LTTP SPS-3 test section A350 and A340 control section in New York.

Although a treated pavement section may have had three data points collected before treatment and/or after treatment, it may or may not have been accepted for analysis. The before treatment and/or the after treatment time-series data of some of these test sections indicated that the pavement condition and/or distress was improving over time without the application of any treatment, as shown in figure 49.

1 inch/mi = 0.0158 m/km.

Figure 49. Graph. IRI versus elapsed time for LTTP SPS-1 test section 0119 in Texas.

In the absence of a pavement treatment, most pavement sections deteriorate over time. When the pavement condition and/or distress data indicated improvement over time, without the application of treatment, the data precipitated negative parameters of the pavement performance model (negative slope) (i.e., improved condition and/or distress without treatment). Such pavement condition and distress trends could occur for various reasons including the following:

• Human error and/or inaccuracy while collecting the data. The subjectivity of assigning distress severity levels and estimating the extent of the distress could generate inaccuracies in the time-series data. Properly calibrated sensor measured data would not exhibit this problem because no human subjectivity was involved.
  
  
• Data inaccuracy due to the employed equipment, such as calibration, malfunction, or changing equipment type between surveys over time.
  
  
• Environmental conditions from one data collection cycle to the next. For example, the crack opening was wider on cold days than on warmer days as a result of thermal expansion and contraction. This could change the assigned crack severity level and/or the observance of the length of the cracks. Likewise, temperature differential could cause curling on certain days, which would influence the measured pavement roughness in terms of IRI.

When the pavement condition and/or distress showed improvement over time without any treatment application, the data yielded infinite RFPs and/or RSPs, and the pavement performance could not be assessed. For example, the data in figure 49 and the associated exponential equation indicate improvement of the IRI over time and a negative exponential power component. That is, according to the given data and the equation, the IRI would never reach the threshold value, and therefore RFP was infinite, which was not practical. Consequently, any time-series condition or distress data showing improvement over time without the application of treatment was not included in the analyses of the pavement performance.

Severity Level

Finally, the analyses of cracking data in this study were based on the sum of the data for low, medium, and high severity levels.

SUMMARY AND CONCLUSIONS

This chapter presented the various data elements of the more than 2,500 test sections included in the LTPP Program that were downloaded from the 6 data volumes housed in the LTPP database Standard Data Release 28.0. The data were organized in a special format and readied for analyses. In addition, the pavement management databases from three State transportation departments—CDOT, WSDOT, and LADOTD—were requested and received. From each database, several pavement projects that were subjected to certain treatments in the past were identified, and their data were downloaded from the respective databases and formatted for analyses. Based on thorough and exhaustive review of the data, the research team reached the following conclusions:

• At the time of this study, 800 test sections were still active, whereas 1,700 were decommissioned.
  
  
• The manually collected pavement distress data were more numerous and consistent than the semi-automated data. The manual data were used in the analyses of the structural pavement performance of each test section.
  
  
• The database for some of the test sections contained only two data points over time before and or after treatment. For those test sections that were subjected to overlay or mill-and-fill treatments, an initial data point was added to rut depth and to each cracking type data.
  
  
• The addition of one initial data point was based on engineering logic and increased the number of test sections that could be analyzed.
  
  
• A total of 2,674 treatments were applied to 1,301 test sections of the 1,555 test sections included in the SPS-1 through -7 and GPS-6, -7, and -9 experiments.
  
  
• For each treated test section, the data were organized for the analyses of pavement performance before, after, or before and after treatment, depending on the data availability.