Optimization of Traffic Data Collection for Specific Pavement Design Applications

Chapter 6. Define Traffic Collection Requirements

From a pavement design point of view, traffic data collection requirements need to be established as a function of the tolerable error and the desirable confidence in NCHRP 1-37A design guide life predictions. The error is defined as the difference between pavement life predictions obtained under a particular scenario and those obtained under true traffic (scenario 1-0). This error, expressed in percentage, was determined from the series of pavement life prediction plots presented in Appendix D. It comprises two components, as shown schematically in figure 17, which are labeled as "A" and "B." "A" is the estimated error from the traffic input of a continuous scenario or the mean traffic input of a discontinuous time-coverage scenario. "B" is the additional error possible in discontinuous-coverage scenarios by considering the lowest percentile input for all traffic input estimates simultaneously. Although doing so is very conservative, it allows establishment of the statistical maximum error in predicting pavement life, and therefore, it answers the question of reliability, as explained earlier.

Figure 17. Components of the percentage difference between pavement life predictions for scenario X and those for scenario 1-0.

Statistics for quantity "A" were computed by scenario type using the NCHRP 1-37A design guide performance predictions for the sites analyzed. These results are plotted by pavement type and traffic level in figures 18 through 21. In interpreting these results, consideration was given to the source of traffic input for each scenario, as outlined in table 22. To facilitate interpretation, the scenarios in these figures were arranged in three groups:

• SS AVC, SS, or R WIM, which indicates site-specific vehicle classification distribution data combined with site-specific or regional axle-load distribution data.
• R AVC, R WIM, which indicates regional vehicle classification distribution data combined with regional axle-load distribution data.
• R or N AVC, N WIM, which indicates regional or national vehicle classification distribution data combined with national axle-load distribution data.

Analysis of figures 18 through 21 reveals the following information:

• For scenarios 1-1 through 2-3 (i.e., site-specific AVC, site-specific or regional WIM), the mean errors are relatively small and, in general, increase as site-specific information and time coverage decreases. They range from -10.8 percent to +9.9 percent.
• For the remaining scenarios (i.e., 3-0 through 4-7), the mean errors are considerably larger. This is not surprising because they rely increasingly on regional or national vehicle classification and axle-load distribution data.
• The sign of the mean errors seems random (i.e., some scenarios underpredict pavement life, while others overpredict it). For scenarios 1-1 through 2-3, this largely depends on whether the estimated MAFs resulted in higher or lower seasonal damage accumulation, respectively, than the actual MAFs. For scenarios 3-0 through 4-7, this depends on whether the regional or national vehicle classification and axle-load distributions used were heavier or lighter, respectively, than the actual ones. Given the random nature of the signs of these mean errors, it is anticipated that if a considerably larger sample of pavement-performance predictions was available, the mean errors for all of the scenarios would become negligible.
• Overall, it is observed that the standard deviation in the errors decreases as the truck traffic volumes increase (i.e., the AADTT > 800 and AADTT > 1,200 sections exhibit lower error variation than the AADTT = 800 and AADTT = 1,200 sections, respectively). This is not surprising, considering that short-term sampling yields more accurate AADTT estimates where traffic volumes are high.
• For scenarios 1-1 through 2-3, the standard deviation in the errors increases as site-specific information and time coverage decreases. In general, it ranges from 1.8 percent to 23 percent.
• For the remaining scenarios (i.e., 3-0 through 4-7), the standard deviation in the errors is considerably higher and, in general, it increases as site-specific information and time coverage decreases. In general, it ranges from 14.3 percent to as high as 70 percent.

Statistics for pavement prediction error component "A" were computed for all 17 sites analyzed (table 29).

Indeed, the mean values of this error component are negligible. Ranges in these errors were estimated as a function of the selected level of confidence (i.e., 100 percent minus the probability of exceeding that error), assuming that the mean error values are zero. Because the sample size is small, the Student's t-standard deviate was used. The values corresponding to one-sided probabilities of 75 percent, 85 percent, and 95 percent for 17 observations are 0.69, 1.07, and 1.74, respectively. The error ranges obtained as a function of the probability of exceeding them are listed in table 29 and plotted in figure 22.

Figure 18. Statistics for error component "A" in life predictions (percent), flexible pavement sites with AADTT = 800 trucks/day/lane.

Figure 19. Statistics for error component "A" in life predictions (percentage), flexible pavement sites with AADTT > 800 trucks/day/lane.

Figure 20. Statistics for error component "A" in life predictions (percentage), rigid pavement sites with AADTT = 1,200 trucks/day/lane.

Figure 21. Statistics for error component "A" in life predictions (percentage), rigid pavement sites with AADTT > 1,200 trucks/day/lane.

Figure 22. Estimated range in NCHRP 1-37A design guide pavement life prediction errors from mean traffic input.

Assuming mean traffic input, figure 22 can be used to establish the least effort traffic data collection scenario that will provide a maximum acceptable pavement life prediction error under a selected level of confidence. For example, scenarios 3-0, 3-1, and 4-2 are among the least effort scenarios capable of a 25-percent maximum error in predicting pavement life (e.g., 3.5 years in a 14-year design period) with a 75-percent confidence. A better traffic data collection scenario would be needed to either decrease the level of the acceptable error or increase the confidence that it will not be exceeded.

The error ranges described above were obtained assuming mean traffic input for each discontinuous traffic data collection scenario. Additional errors were computed because of the variation in traffic input resulting from the sampling scheme used within each of the traffic data collection scenarios that involved discontinuous time coverage. These reflect the error component "B" defined in figure 17. As described earlier, this error component in predicting pavement life was computed from the NCHRP 1-37A design guide life estimates obtained by inputting the low percentile for all of the traffic elements simultaneously. Table 30 shows the mean and the standard deviation of these errors by traffic input percentile level. It also shows the standard deviation in the mean errors computed by dividing the standard deviation of the errors by the square root of the number of degrees of freedom (i.e., =4), according to the central limit theorem. Table 31 gives the ranges in the mean error component "B" by traffic input percentile level. These were computed from the data in table 30 by summing the mean error plus the product of the standard deviation of the mean error multiplied by the Student's t-deviate (i.e., 0.69, 1.07, and 1.74, as described earlier).

The combined range in the two error components "A" and "B" was computed by percentile level by adding the range in the error component "A" to the range in the mean of the error of component "B." The results are shown in table 32 and plotted in figure 23. Figure 23 was compiled assuming that the lowest percentile of all of the traffic input for a discontinuous-coverage scenario could be input simultaneously during design. As mentioned earlier, this is very conservative; however, it addresses the question of reliability to guarantee the designer that a particular error level will not be exceeded given a level of confidence.

Assuming low-percentile traffic input, figure 23 can be used to establish the least-effort traffic data collection scenario that will provide a maximum acceptable pavement life prediction error under a selected level of confidence. Compared to the earlier example, scenario 3-0 is the only one among the least effort scenarios identified earlier capable of a 25-percent minimum error in predicting pavement life with a 75-percent confidence. A better traffic data collection scenario would be needed to either decrease the level of the acceptable error or increase the confidence that it will not be exceeded.

The main observations drawn from figure 23 are summarized below:

• Discontinuous traffic data collection scenarios involving site-specific WIM data (scenarios 1-1 and 1-2) are inferior to continuous-coverage, site-specific AVC data (scenario 2-0). This is because partial WIM coverage does not yield site-specific MAFs, which are necessary for accurately modeling seasonal damage in the NCHRP 1-37A design guide.
• Scenario 2-0 is capable of predicting pavement life with an error lower than 10 percent, 16 percent, and 27 percent for confidence levels of 75 percent, 85 percent, and 95 percent, respectively.
• Where continuous site-specific truck counts are combined with regional load and classification data (scenario 3-0), life prediction errors may range from 25 percent to 64 percent, depending on the desired confidence level.
• Where continuous site-specific truck counts are combined with regional classification and national load data (scenario 4-0), life prediction errors may range from 27 percent to 68 percent, depending on the desired confidence level.
• Where continuous site-specific truck counts are combined with national axle-load and classification data (scenario 4-4), life prediction errors may range from 30 percent to 76 percent, depending on the desired confidence level.

Figure 23. Estimated range in NCHRP 1-37A design guide pavement life prediction errors from low-percentile traffic input.