Optimization of Traffic Data Collection for Specific Pavement Design Applications

Executive Summary

This study presents a comprehensive approach for establishing the minimum traffic data collection effort required for pavement design applications satisfying a maximum acceptable error under a prescribed confidence level. This approach consists of simulating the traffic data input to the new National Cooperative Highway Research Program (NCHRP) 1-37A design guide for 17 distinct traffic data collection scenarios using extended-coverage, weigh-in-motion (WIM) data from the Long-Term Pavement Performance (LTPP) database. This simulation involves data typically collected by other technologies, such as automated vehicle classification (AVC) and automated traffic recorders (ATR). These scenarios are described in table 8 in the report.

Extended coverage was defined as 299 or more days per year of level E WIM data, (i.e., data that has passed the quality control (QC) checks conducted by State departments of transportation (DOTs) and the LTPP regional support contractor offices). Analysis of LTPP Standard Data Release 16.0 revealed a total of 178 general pavement studies (GPS) satisfying this requirement. For all of these sites, central traffic database (CTDB) data were extracted in the form of daily summaries (level 3). From these sites, a total of 30 sections (15 flexible and 15 rigid) were selected for NCHRP 1-37A design guide pavement performance estimation. The selection was based on the widest possible distribution of average annual daily truck traffic (AADTT) volumes and structural thicknesses.

A number of the traffic data collection scenarios simulated involved continuous site-specific data coverage for axle loads, classification, or counts, while others involved discontinuous site-specific data coverage (1 month per season, 1 week per season, and so on). Data elements, which were assumed to be unavailable at a site for simulation purposes, were estimated from regional data. Regional vehicle classification and axle-load data were obtained from the remaining LTPP sites identified using clustering techniques. Scenarios involving national data used the default traffic input in the NCHRP 1-37A design guide. For each of the traffic data collection scenarios involving discontinuous coverage of site-specific data, statistics for each traffic data element were computed by considering all possible time-coverage combinations. This allowed the establishment of low percentiles for each input to simulate underestimation of the actual traffic volumes/loads at a site. This was considered to be critical since it would result in thinner pavement designs that failed prematurely. Four confidence levels were selected: 75 percent, 85 percent, 95 percent, and 99.9 percent. Traffic input for the continuous-coverage traffic data collection scenarios involved no variation because of the sampling scheme used. All scenarios were simulated using a 4 percent annual growth in AADTT. Additional analysis was conducted to compute the annual growth rate in AADTT and its effect on pavement life predictions.

The NCHRP 1-37A design guide pavement life predictions for each scenario were analyzed to compute percentage errors in pavement life predictions with respect to the life predictions obtained under continuous site-specific WIM data (scenario 1-0). Reasonable life predictions were obtained for 17 of the 30 sections analyzed; the remainder experienced either premature failures or no failure at all. Two pavement life prediction error components were identified:

• "A" is the estimated error from the traffic input of a continuous scenario or from the mean traffic input of a discontinuous time-coverage scenario.
• "B" is the additional error possible in discontinuous-coverage scenarios by inputting the lowest percentile input for all traffic input estimates simultaneously.

Computing statistics for error component "A" for all 17 sections revealed that its mean is negligible for all of the scenarios analyzed. Its standard deviation allowed establishment of a range of errors by confidence level (table 1 below and figure 22 in the report).

Statistics for error component "B" were processed to yield the mean error and the standard deviation in the mean error by traffic data collection scenario. This allowed computation of the range in mean error resulting from specifying the lowest percentile for all traffic inputs simultaneously. While this is quite conservative, it addresses the question of reliability, so that the designer is guaranteed that given a level of confidence, a particular error level will not be exceeded. Overall error was computed by adding the range in error from component "A" to the range in mean error from component "B." The results were plotted in a three-dimensional plot, indicating the maximum error by confidence level for each of the traffic data collection scenarios analyzed (table 2 below and figure 23 in the report). Figure 23 can be used to establish the minimum required traffic data collection effort, given the acceptable error and the desirable level of confidence.