Study of LTPP Pavement Temperatures

CHAPTER 3. ERRORS FOUND AND RESPONSE

This chapter is divided into the types of errors found using the temperature plot scanning procedure and variations of errors within the types.

IRRELEVANT IR TEMPERATURES

Many of the records in the M04 tables, for the testing on SPS-1, SPS-2, and SPS-8 projects, have entries in the PVMT_SURF_TEMP field when the tests were on the subgrade, granular base, lean concrete (lean PCC), or permeable asphalt treated base (PATB); therefore, many data are likely to be incorrect. The IR data on subgrade and granular base are not relevant, and all should be replaced with nulls. A determination should be made regarding the relevance of the temperatures that are available on lean PCC and PATB

FWDs WITHOUT AN IR SENSOR

Some testing was done with an FWD equipped with an IR sensor that was either missing or not working. In either case, the filter program that processes the field files into the database reads the blank field as a zero ("0"), a reading that is incorrect. In addition, the filter program should be modified to check for lane designations—"S" (subgrade), "G"(granular base), "L" (lean PCC), and "P" (PATB). All these records should have the value in the PVMT_SURF_TEMP field replaced with a null or blank field.

FWDs without IR sensors typically exhibited nonsensical values in the header file for calibration numbers, in PVMT_SENSOR_VOLTAGE_0C and PVMT_SENSOR_VOLTAGE_100C fields of the M02 tables. The M04 tables can be searched for the records that correspond to configurations that have nonsensical values. These cases will be easy to identify. These records should then be evaluated because some FWD operators were able to enter surface temperatures from a hand-held IR thermometer. Records that did not have IR sensors were identified during the temperature plot scans rather than the above-described method. This can be easily written into the FWD filter program and also used as an independent check for bad IR data.

Table 4 lists examples of sites where IR data were in the database, but the data were not valid. These cases are a sample of such cases for Region 2. Note that these may not include all of the data; they are only part of the data for the day. An example is Section 271028 on May 10, 1994, where the IR stopped working after the test at 13:45, as shown in the example in figure 8.

Figure 8. Graph. Example of an IR sensor ceasing to function.

TIME ENTRY ERRORS

A number of sections have errors in time entry. They come in a variety of forms, including the time of the manual measurement; time zone time errors for either the manual recordings or FWD computer time, or both; daylight savings time errors; or simple data recording or entry errors.

Time of Manual Measurement

The most noticeable form of this error is recording and entering a manual temperature measurement that was made in the afternoon as the time based on a 12?hour clock, such as recording 13:00 as 1:00. These errors, found while scanning the daily temperature plots, were readily noticeable. Screening of manual temperature times recorded between midnight and 6 a.m. would catch most of these errors. Some caution is advised because some testing in Region 4 was at night.

Incorrect Computer Clock Setting

This data entry error, caused by setting the computer clock in the afternoon on 12 hour time instead of 24-hour time, resulted in deflection time stamps that are 12 hours late. In addition, date errors occurred if the first error was not corrected by the next morning. Figures 9 and 10 show examples of such errors, with the additional complication in figure 9 that the shift occurred from 9:00 to 21:00 rather than at noon as shown in figure 10.

Figure 9. Graph. Time- temperature plot showing computer time error.

Figure 10. Graph. Time-temperature plot showing 12-hour computer time error.

Time Zone Errors

This error could have occurred when the time of the manual temperature measurement was recorded if the operator did not account for crossing a time zone, or if the operator did not adjust the computer clock for the time zone change. If the operator did not account for the time zone change in the computer or adjust for it while recording the manual temperatures, the error generally is undetectable. This error probably is not too serious because it likely is consistent with the rest of the data.

Daylight Savings Time Errors

These errors are similar to time zone errors, except they occurred only during the first few weeks of April or the first few weeks of November. As with time zone errors, if both the manual and FWD times go unadjusted, the error is not easily detectable and probably is not too serious because it likely will be consistent with the rest of the data. Double entries of FWD data can occur with either the time zone or daylight savings time errors if the region filters the data into IMS, recognizes the problem, editsthe field files to correct the time, and then refilters the FWD file into the database without removing the initial entry. Occurrences of this error were found during both the temperature scans and deflection checks. (Deflection checks are a separate process that is not covered in this report.) For the most part, this problem has been identified and the Region must now review the data to determine which set is correct and corresponds to the deflection data set as corrected and reported previously.

Simple Time Recording and Entry Errors

These errors are found during scanning of the time temperature plots as misplaced times. If the error is large enough and there is sufficient temperature change, incorrectly recorded or entered time will stand out. Suspect data can be identified, but it requires that the region review the data sheets and make appropriate corrections.

TEMPERATURE ERRORS

There are a number of temperature data recording or entry errors. Errors such as transposing the numbers, incorrectly entering the tens-place value, and reversing the temperature holes were identified by visually scanning the time?temperature plots. These errors can be expected; and as expected, they did occur. If the errors are sufficiently large, they can be detected by observing the plots, as shown in figure 11. Generally, the smallest detectable error is -12 °C (10 °F), as shown in figure 12. Note that the LTPP protocol required the temperature measurements to be in Fahrenheit.

An alternative process that could be used to search for temperature errors is to use the BELLS2 method to estimate the temperature at the depth used in the database. This procedure could be used for all flexible pavements where the previous day's average air temperature data are available. The IR calibration problems described earlier first need to be resolved. Then the IR temperatures can be used to estimate in?depth temperatures, which, in turn, can be compared to the measured temperatures. That process might detect errors smaller than those identified in visual scans; however, this is uncertain.

The BELLS2 equation can also be compared to temperatures in rigid pavements. Preliminary calculations (before data cleaning) were carried out for both surface types (AC and PCC), and the results were encouraging. Figures 13 and 14 show the results for FWD SN 8002-058 while equipped with a Raytec IR sensor. Note that there are many outliers that may be a result of data problems such as infrared extrapolations and manual temperature data errors. As seen, the data band in figure 13 is much tighter than in figure 14; figure 13 data is for manual 1 data, generally from about 25 mm below the surface, and figure 14 is from manual 3, which generally are 25 mm above the bottom of the bound layer, and therefore, more likely to have greater variation. The adaptation of BELLS2 (or BELLS3) to PCC surfaces could be developed from this data and would be another LTPP product. Time was insufficient to complete that as part of this project, but the concept has been tested with encouraging results.

Figure 11. Graph. BELLS2 prediction for manual 1 temperatures using default IR data before cleaning

Figure 12. Graph. Minimum detectable error possible by visual scan.

Figure 13. Graph. Manual 1 and BELLS2 compared (before data cleaning; all surfaces).

Figure 14. Graph. Manual 3 and BELLS2 compared (before data cleaning; all surfaces

SUMMARY OF ERRORS DETECTED BY REGION

The temperature data were scanned by Region, one date at a time. Because of time and budget restraints, the scanning was complete for only Regions 2 and 4, and about 75 percent complete in Region 1. Region 3 temperature data were not scanned at the time this report was prepared. In Regions 2, 3, and 4, where SPS-1 and SPS-2 testing was done on the subgrade (S), granular base (G), permeable asphalt treated base (P), or lean concrete base (L), no manual temperature measurements were made on these surfaces. Following is a list of the number of individual section-days available for scanning:

• Region 1: 1,247 section—days-total includes no P, L, S, or G section-days.
• Region 2: 1,995 section—days-including 12 P, 4 L, 41 S & 20 G section-days.
• Region 3: 2,188 section—days-including 0 P & L, 24 S & 13 G section-days.
• Region 4: 2,078 section—days-including 21 P, 3 L, 92 S & 41 G section-days.

Region 2

The following paragraphs summarize the Region 2 temperature problems found in scanning the plots.

Sections with Missing Temperature Data. In all of the sections of test data in Region 2 that have reached Level E, the sections shown in table 5 have no temperature data. An approximation of pavement temperatures could be made based on historic climatic data specific to this SPS-5 site or similar sites nearby and other test dates with IR and manual temperature. These approximations (predictions) come within about ±5 °C. Other sections also have missing data in the IMS.

Sections with Missing Manual Data. There could be several reasons why a section has no manual data:

• Testing was on subgrade, granular base, PATB, or lean concrete; manual temperature measurements are not required for these surface types
• Manual temperatures were not measured.
• Manual temperatures were measured but not entered into the database
• Manual temperatures have not advanced to Level E
• Minnesota Test Road (MnROAD) GPS sections—manual temperatures were not measured. All of the sections at MnROAD were instrumented with thermocouples. The temperature data from MnROAD need to be transferred into the IMS, or users need to be directed to where the data is available on the MnROAD web site.

The first group consists of tests conducted on SPS-1, SPS-2, or SPS-8 during construction. These sections did not require manual data because the tests were conducted either on subgrade, granular base, or PATB; however, some of these sections did have IR data. The IR data were not screened and no assessment is given as to the validity of these data. It may be important to screen the PATB data because the deflections on this material are temperature dependent. Manual temperatures were not made on PATB because of the permeability—any holes drilled in the material would not hold the requisite heat transfer fluid. It is recommended that the IR data on PATB be screened for suitability.

The next three groups cannot be individually identified at this time; however, sections with missing manual data are identified. A list of the sections with missing manual data should be submitted to the four Regions, which could then check their records to verify that the data are missing or that the data exist but have not been entered (in which case the data can be entered).

Sections that still do not have manual data, but do have IR data, could have a set of computed temperatures for the pavement based on the BELLS2 equation for asphalt surfaces. Concrete surfaced sections could be similarly treated following an evaluation of the BELLS2 equation for use on PCC surfaced pavements. The calculated temperatures could be approached in at least two ways:

• Apply BELLS2 to cleaned and calibrated IR data.
• Calculate the in-depth temperatures after calculating a new set of coefficients for the IR sensor and in-depth temperatures for specific calibration periods for each specific IR sensor. Using the new set of coefficients, calculate the in-depth temperatures for sections with missing data. Usually the best procedure to accomplish this is to use some weighting of site-specific estimated parameters with globally estimated parameters; we cannot identify these factors without further, more time-intensive analyses.

Sections with Apparent Manual Temperature Errors. Apparent manual temperature errors appear in 49 section-days; visual inspection of temperature plots makes the errors seem obvious. Of the 49 section-days, one has IR and manual data that are both suspect, and the rest have specific data elements identified as suspected errors. Identified errors should be checked by the Regions. A correction can be made when the error is an entry error. When the error may have been a recording or reading error, a decision must be made to either remove the data, leave it as is, or change the number to the expected value. A good method to handle the changed values is with computed parameters held in separate tables. On the other hand, separate tables could be cumbersome for researchers. To create the best set of data for researchers, it is possible to enter a separate data set containing both actual and computed data.

Section-Days With Possible Time Errors. There are 34 section-days with possible time errors. There are several ways time errors can be made:

• Recording afternoon times using a 12-hour clock system rather than the protocol 24-hour clock system. A time at 1 p.m. time, if entered as 1:00, is recorded in the database as 1:00 a.m.
• Not adjusting for going off or on daylight savings time in the computer clock or manual time recordings.
• Not adjusting, or incorrectly adjusting, for time zone changes.
• Filtering in FWD data twice, once with the original data, and once with the time adjusted. There are four section-day file cases of this.

Region 4

The following paragraphs summarize the Region 4 temperature problems caused by scanning the plots.

Sections with No Temperature Data. For all of the sections of test data that have reached Level E, the sections listed in table 6 have no temperature data at all. All of these sections are in Alaska; the temperatures for the first two sections probably were never measured, and the last three may not have been entered in time to be part of this analysis. An approximation of pavement temperatures could be made based on historic climatic data specific to these sites or similar sites nearby and other test dates with IR and manual temperature. These approximations (predictions) come within about ±5 °C.

Sections with Missing Manual Data. In Region 4, aside from the testing on the unbound layers, PATB, and lean PCC, manual temperature data are missing from 144 sections, not including all the SPS-3s and SPS-4s. Following is a list of grouped sections (sections here refer to section-days) missing manual temperature data:

• 43 flexible GPS sections.
• 14 rigid GPS sections—13 jointed and one of continuously reinforced concrete pavement (CRCP).
• 51 SPS-5 sections.
• 31 SPS-6 sections.
• 5 SPS-8 sections.
• All SPS-3 and SPS-4 sections.

There could be several reasons why these sections have no manual data:

• Manual temperatures were not measured.
• Manual temperatures were measured but were not entered into the database.
• Manual temperatures have not advanced to Level E.

The three groups could not be individually identified at the time of the study; however, it was possible to identify sections with missing manual data. A list of the sections with missing manual data should be submitted to the regions. The regions should then check their records to verify that the data are missing, or that the data exist but have not been entered, in which case the data can be entered.

Sections that still have no manual data, but that do have IR data, could have a set of computed temperatures for the pavement based on the BELLS2 equation for asphalt surfaces. Concrete surfaced sections could be similarly treated following an evaluation of the BELLS2 equation for use on PCC surfaced pavements. The calculated temperatures could be approached in at least two ways:

• Apply BELLS2 to cleaned and calibrated IR data.
• Calculate the in-depth temperatures after calculating a new set of coefficients for the IR sensor and in-depth temperatures for specific calibration periods for each specific IR sensor. Using the new set of coefficients, calculate the in-depth temperatures for sections with missing data. Usually the best procedure to accomplish this is to use some weighting of site-specific estimated parameters with globally estimated parameters; we cannot identify these factors without further, more time-intensive analyses.

Sections with Apparent Manual Temperature Errors. There are 82 section-days with apparent manual temperature errors. The errors seem obvious by visual inspection of temperature plots. These errors should be checked by the Region. In the case the error is a data entry error, a correction can be made. In case the error may have been a recording or reading error, a decision must be made to either remove the data, leave it as-is, or change the number to the expected value. Changed values may best be treated similar to computed parameters (i.e., they may be held in separate tables); however, this would be cumbersome for researchers. To create the best set of data for researchers, a separate data set that contains both actual and computed data should be made.

Section-Days with Possible Time Errors. There are 104 section-days with possible time errors. There are several ways time errors can be made:

• Recording afternoon times in the 12-hour clock method rather than 24-hour clock as per protocol. A 1:00 p.m. time, if entered as 1:00, is recorded in the database as 1:00 a.m.
• Not adjusting for going on or off daylight savings time in the computer clock or manual time recordings.
• Not adjusting, or incorrectly adjusting, for time zone changes.
• Filtering in FWD data twice, once with the original data and once with the time adjusted. There are four cases of this.
• Night testing where all the tests taken are listed for the day the testing was completed. For example, testing started at 22:00 or 10:00 p.m. and ended 2:00 a.m. The tests between 22:00 and midnight will be associated with the following day. This is a data filtering and field program operation problem.

The possible errors have been specifically identified and feedback reports will be submitted. This is an uncommon problem; no solution is offered at this time other than to manually edit the dates.