U.S. Department of Transportation
Federal Highway Administration
1200 New Jersey Avenue, SE
Washington, DC 20590
Federal Highway Administration Research and Technology
Coordinating, Developing, and Delivering Highway Transportation Innovations
|This report is an archived publication and may contain dated technical, contact, and link information|
Publication Number: FHWA-RD-02-071
Date: March 2005
Study of LTPP Pavement Temperatures
CHAPTER 2. DATA EXTRACTION AND PROCESSING
The approach to evaluating the quality of the manual and infrared temperature data was to compare the data sets and individual data elements to previous and subsequent data elements. Two forms of temperature measurements are available for the comparison-(1) infrared surface pavement data from table MON_DEFL_LOC_INFO (M04) and (2) manual in-depth pavement data from tables MON_DEFL_TEMP_DEPTHS (M21) and MON_DEFL_TEMP_VALUES (M22). The two sets of data, IR and manual, are then compared to each another. The calibration settings for the infrared sensors are contained in the table MON_DEFL_DEV_CONFIG (M02). The data fields used for the project are listed in the Appendix.
The dataset that had reached "Level E" status was furnished on CD-ROM in ASCII format. The temperature data and other relevant data were extracted from the tables and assembled into individual section files, uniquely identified by STATE_CODE and SHRP_ID. The infrared and manual temperature data were merged into various files, one for each section, called SITESM files. The data in these files are sorted by date and time of test. Three additional data fields were included in these files, one calculated field for the infrared temperature based on the manufacturer's default factory calibration factors, one for the manufacturer of the infrared sensor, Williamson (W) or Raytec (R), and one for the previous day's average air temperature, provided by an FHWA data contractor for all the GPS sites. (The previous day's air temperature can be used to predict temperatures in the pavement based on the BELLS2 prediction model.) During the assembly of each of these files, an interpolated infrared surface temperature was calculated for each manual temperature record using standard linear interpolation (or extrapolation) methods. The extrapolated values were less reliable.
The infrared sensors used on the FWDs generate an electrical response that is converted to an electrical potential (in millivolts), which is linearly related to the surface temperature of the pavement. The two IR devices each have two default millivolt (mV) values that correspond to the sensor output when the pavement surface temperature is 0 °C and 100 °C. Table 1 contains the default calibration values for the two types of sensors used.
These default values are included in the header file of the FWD field program, which uses these values to linearly interpolate (or extrapolate) the IR sensor response to the surface temperature. Figure 1 gives the equation to convert millivolts to temperature.
Figure 1. Equation. Converting millivolts to temperature.
For example, if the output from the Williamson sensor is 1440 mV (half way between 800 and 2080 mV), the surface temperature is 50 °C.
LTPP protocol for the FWD required periodic calibrations of the IR sensors. A water-and-ice mixture was used for a 0 °C reference, and nearly boiling water was the hot temperature reference. (Boiling water as a 100 °C reference proved to be impossible to use for calibration because of the interference of the steam and different emissivity.)
The process of calculating the default IR values is to use the equation in figure 1 and calculate the actual IR sensor output in millivolts. Using IMS data field names, figure 2 gives the equation in figure 1 solved for millivolts.
Figure 2. Equation. Calculate default IR values to derive IR sensor outputin millivolts
The three requisite input variables to this equation are available in the data tables. After the millivolt value is calculated from the equation above, the factory default calibration values and the millivolt output are used in the equation in figure 1 to calculate the default IR reading for T. This procedure was done for all of the data evaluated, and the calculated default IR (D.IR) values are included as a separate field in the combined files.
IR Calibration Issues
Lukanen et. al.,(3) in a previous LTPP project found some significant differences between the various infrared sensors mounted on the various LTPP FWDs. The calibration protocols resulted in more variation from sensor to sensor than would result if the factory calibrations had been used. As a result of these findings, the infrared readings, as calibrated by the regions, and the manufacturers' default calibrated infrared readings were regressed against the measurements in the LAYER_TEMPERATURE1 field. The results for the North Atlantic and Western Regions appear in tables 2 and 3. (LAYER_TEMPERATURE1 is typically measured in a 25-mm deep hole with 5 mm to 10 mm heat transfer fluid at the bottom of the hole.) The manufacturers' calibrations resulted in more consistency from year to year and sensor to sensor.
Comparing the manual and infrared temperatures provides a means to scan for errors. For example, the small crosses in figure 3 are D_IR pavement surface readings; the circles are the manually measured temperatures at about 25 mm below the pavement surface for one of the LTPP FWDs. The plot shows that the IR sensor did not work well during the winter, and it became erratic after May 24, 1995. For each of the IR calibration periods, the comparisons were used either to find times when the equipment malfunctioned or to find errors in the upper manual measurements at 25 mm depth.
Figure 3. Graph. Infrared and manual temperatures for SN 8002-129.
A key premise to this exercise is the idea that the manual temperature is considered to be the best representation of the pavement temperature. The temperature meters and probes that measure the temperature in the heat transfer liquid placed at the bottom of each manual temperature measurement hole were not subject to any rigorous calibration or verification. The manufacturer's specifications and certifications for the meters were accepted as statements of their accuracy; however, it is easy enough to check the probes and meters against reference thermometers that are traceable to the National Institute of Standards and Technology (NIST).
Two models of infrared sensors were used over the course of the LTPP project. The initial IR sensors were manufactured by Williamson. The first set of four FWDs were SHRP FWDs. A second set of four FWDs was delivered in 1995, one to each of the Regions. These machines were equipped with Raytec sensors. During the summer of 1995, the original Williamson sensors were replaced with Raytec sensors. It was also found that the factory default calibration settings for the Raytec sensors were more consistent from sensor to sensor than the calibration settings for the Williamson sensors. The Williamson IR sensors tended to correlate well (although not necessarily one-to-one) with the manually measured pavement temperatures on a case-by-case basis; however, there were significant differences from one IR sensor to the next. The Raytec IR sensors were more consistent from sensor to sensor. The regression results shown in tables 2 and 3 confirmed the earlier findings.
Post-Testing of Infrared Calibration Constants
The temperature measurements conducted in LTPP make it possible to consider post-testing calibration of the infrared sensors. If we accept the manual temperatures as a reliable temperature reference, we can use the shallow manual temperatures and the regression results shown in tables 2 and 3 to calculate a new set of calibration factors for 0 °C and 100 °C. This process also requires the selection of a reference intercept and slope. The weighted average of the slope and intercept for the more reliable Raytec sensors, for example, could be used as the reference relationship between the IR sensors and the top manual temperatures. The equation form in figure 4 shows the relationships reported in tables 2 and 3.
Figure 4. Equation. Relationships in Tables 2 and 3.
Replacing the IR term with the equation shown in figure 1, the equation in figure 4 can be rewritten as shown in figure 5, where a is the intercept and b is the slope.
Figure 5. Equation. Rewritten Figure 4 equation to replace IR term.
Converting Cal100 - Cal0 to and M1 - a to M', the equation can then be rewritten as shown in figure 6.
Figure 6. Equation. Using conversion equation to rewrite equation.
By using the two calculated variables, M' as the dependant variable and mV as the independent variable, all of the output for any of the sensors and dates can be regressed. The new regression intercept would be (-100*b*Cal0) / () and the slope would be (100*b) / (), yielding two equations with two unknowns. can be calculated by = (100*b) / slope. Cal0 can be calculated by Cal0 = (Int. * ) / (-100*b). This process can be applied to each individual calibration period for the Williamson IR sensors-and to a Raytec IR sensor, for that matter, if it correlates well with M1, but it has significantly different regression coefficients.
Applying this process to the IR sensor on FWD S/N 058A results in a new set of calibration factors, of 778.98 and 2,122.85 for Cal0 and Cal100 respectively. Using the new calibration factors, a new set of computed values (abbreviated in tabular form as "C.IR") can be developed.
The method used to search for errors that was most productive was to manually view time plots of the temperatures measured on a given section for individual test days. This search was automated by using a spreadsheet macro to allow rapid review of the plots of the data by simply clicking on a "spinner bar" to step forward or backwards through the sections, day by day. There was too much variation in the temperatures caused by factors such as cloud cover, rain, or shadow effects. These variations made automated screening of the data with preset or data-determined numerical criteria difficult.
An example of an error is provided by the plot in figure 7. The plot shows two sets of manual temperature data, reportedly from the same site and day of test. One set of data indicates a warmer surface and one shows a cooler surface. The erratic IR plot is from two individual sets of data that are superimposed on the same time scale. Clearly, one set of manual data and one set of IR data are incorrect. In all likelihood, the two sets of data are not from the same LTPP site.
Figure 7. Graph. Example of time plot of temperatures.