Verification, Refinement, and Applicability of Long-Term Pavement Performance Vehicle Classification Rules

Chapter 4. Evaluation of Likely Errors in the Total Traffic Loading Estimate When Using Load Spectra Computed With the LTPP Class Rule set And Truck Volumes From State-Specific Rule sets

Analysis of Expected Errors in Total Traffic Loading

As noted earlier in this report, the use of different classification rule sets changes how some vehicles are classified. When a different classification rule set is used for load spectra development than is used to develop the count of traffic volume by vehicle classification, the computed number of heavy and light axles is altered from the correct value. Whether this change increases or decreases the total load estimate depends on which classification rule set is used for load spectra development and which is used for vehicle volume estimation. It also depends on what the vehicle characteristics are at any given site. In some cases, whether a given classification rule set increases or decreases the total traffic load is primarily a function of the site-specific traffic stream.

For this analysis, the primary objective was to determine the size of errors imposed on traffic load estimates when using load spectra developed using the LTPP classification rule set, but applied at sites where the traffic volume by classification estimate comes from a State-specific classification rule set.

Of the seven State classification rule sets tested in this specific analysis, two (Washington WIM and Florida AVC) produced a Class Ratio that was greater than 1 at all 18 test sites. (That is, they consistently produced a larger traffic load than the LTPP rule set alone would have estimated.) Two additional rule sets (Wisconsin and Missouri) produced Class Ratios greater than 1 at a majority of sites (15 increase/3 decrease and 10 increase/ 8 decrease, respectively.) One classification rule set (California AVC) produced Class Ratios that were less than 1 at all TPF sites, and two rule sets produced Class Ratios less than one at a large majority of sites (2/16 for the Florida WIM and 5/13 for the California WIM rule set).

Table 13 shows the mean and standard deviation of the Class Ratio along with the number of increased and decreased traffic load estimates for each the 18 TPF sites tested for each of the examined State classification rule sets.

If changes are examined at the site level rather than by classification rule set, another trend emerges. At sites where heavy traffic loads occur, very little change in the load estimate occurs when different classification rule sets are used, regardless of which different rule set is used. However, if annual traffic loads are light, the potential for large percentage changes in traffic loading estimates caused by the use of different classification rule sets increases greatly. This can be seen in figure 15.

Table 13. Effect of using State rule set volumes and LTPP load spectra averaged across 18 TPF sites.

WIM = Weight in Motion

AVC = Automatic Vehicle Classification

Note: The equation in the figure describes the best fit line obtained from a linear regression for the data shown. The goodness of fit is given by the R² value.

Figure 15. Graph. Plot of total annual impact load and standard deviation of class ratio.

Further examination of the TPF data shows that the loading computations for sites with very large traffic loads tend to be dominated by high volumes of large, heavy trucks. In the vast majority of cases, the most significant contributors to that load are Class 9 trucks. This can be seen in figure 16, which illustrates the relationship between total load and the percentage of trucks in Class 9 at the 18 TPF test sites.

Figure 16. Graph. Total annual impact load versus percent of trucks in Class 9.

Because Class 9 trucks rarely change classification when alternative classification rule sets are used, sites at which Class 9 dominate the traffic loading do not show much variation in total traffic load when different classification rule sets are applied.

On the other hand, at those sites where a significant portion of load is contributed by smaller trucks (Classes 8 and below), the use of alternative classification rule sets can have a much larger impact. This correlation between high percentages of smaller trucks (also stated as a low percentage of Class 9 trucks) and a high standard deviation of the Class Ratio can be seen in figure 17.

Note: The equation in the figure describes the best fit line obtained from a linear regression for the data shown. The goodness of fit is given by the R² value.

Figure 17. Graph. Standard deviation of class ratio versus percent of trucks in Class 9.

However, figure 17 also shows that Class 9 percentage is not the only factor that influences the size of the error that can be caused by using different classification rule sets for the load spectra and volume by classification inputs. The outlier data point in figure 17 is the Louisiana TPF site. This site has a very high percentage of unloaded Class 9 trucks and an unusually large number of very heavy trucks in Classes 10 and 13. When compared with the LTPP rule set, the Class 10 and 13 trucks change classification under many State rule sets. The combined effect of light Class 9 loads and heavy Class 10 and 13 loads yields a higher than expected change in total traffic load when the effects of different classification rule sets are compared.

Conclusions and Recommendations

The analysis presented above shows that when a stream of traffic data is processed using different vehicle classification rule sets, the truck volume counts and load spectra computed by those two different rule sets can be significantly different in at least some vehicle classes, including Classes 4—8, 10, and 13. As a result, the total traffic load computed using vehicle classification estimates from one device and a load spectra computed from data collected from a second device that uses a different classification rule set, including the LTPP rule set, has the potential to contain some errors. The size and significance of that error is highly variable, changing considerably from one classification rule set to another and even from one site to another. Sites that have a high percentage of Class 9 vehicles are not likely to be significantly affected by these errors.

It is also clear from the analysis that the effect of using the LTPP load spectra and State-specific classification count data on the accuracy of the total traffic load estimate is nearly impossible to predict without actually performing a detailed analysis for each site and each State classification system. Variability in the errors comes from the following two sources:

• Specific differences in the parameters used to classify vehicles that are found in the various State and LTPP classification rule sets.
• Actual percentage of vehicles present at any given pavement design site that fit within the set of vehicles that change classifications when the LTPP rule set is applied in place of the State rule set.

To simply illustrate these factors, assume that the State rule set classifies all cars pulling trailers as trucks, and the LTPP rule set classifies no cars pulling trailers as trucks. The “different parameters” are the ability to identify a car. If a road is restricted to heavy trucks and actually experiences no cars pulling trailers, then this difference in design has no effect on the truck volume estimate. On the other hand, if the road experiences very heavy recreational vehicle traffic, there will be many such vehicles, and the State rule set will badly overestimate the volume of trucks. This overestimation of truck volume when multiplied by the LTPP load spectra will overestimate the traffic load created by that class of vehicles.

Whether this overestimation of load is significant is a function of what other traffic loads are being applied to that pavement. That is, if the only trucks using the road are those that can be “confused” with cars pulling trailers, the total load estimate will be very poor. On the other hand, if the road carries a large number of trucks in other categories that are not affected by the choice of classification system (such as classic five-axle tractor semi-trailers), then the overall load estimate is likely to be reasonably accurate, despite the errors caused by using two different classification rule sets.

Application of the LTPP Load Spectra Collected Using the LTPP WIM Rule Set at Other Pavement Analysis Sites

The implications of these findings for the application of load spectra developed using the LTPP classification rules are fairly straightforward. It should be possible to use LTPP load spectra at other LTPP test sites without the risk of introducing significant error in the traffic load calculation whenever one of the following conditions occurs:

• High traffic loadings (e.g., total annual impact loading estimate is more than 300,000) are present.
• Greater than 60 percent of trucks in Classes 4 to 13 are in Class 9. ^[10]

Sites with annual traffic loading above 300,000 should experience less than 4-percent additional error (with a 95-percent level of confidence) in their total annual impact loading estimates as a result of the use of two vehicle classification rule sets (the LTPP rule set plus a State rule set) in the development of their load estimate. Even with total annual impact loading estimates below 300,000, the added error caused by the mismatch in vehicle classifications from the use of different AVC and WIM rule sets should be below 12 percent (with 95-percent confidence), if Class 9 trucks represent more than 60 percent of all truck volume (i.e., 60 percent of volume in Classes 4 through 13).

At LTPP test sites that fall outside the above parameters (i.e., have lower total annual impact loading estimates and less significant Class 9 truck volume percentages), the use of the LTPP TPF load spectra can result in more significant errors in the estimated traffic load. The data available for tests in this project suggest that at these sites, the 95th percentile error bounds in annual pavement loading estimate (resulting from use of different classification rule sets) ranges between 10 and 34 percent.^[11] In general, the greater the percentage of total traffic load owing to truck Classes 5 and 8, the greater the error will be when using load spectra created with the LTPP rule set.

Based strictly on the computation of the annual traffic loading estimate, it is therefore not recommended to use load spectra from the LTPP TPF sites collected using the LTPP classification rule set in conjunction with pavement study sites that have at least one of the following:

• Total annual impact loading estimates below 50,000 (unless a very high percentage of those loads come from Class 9 trucks).
• Less than 50 percent of truck volume in Class 9 among Classes 4 through 13.

This recommendation applies unless additional work is performed that shows that the specific State classification rule set used to collect truck volume data does not create significantly different truck volumes in truck Classes 5 and 8 than the LTPP rule set when applied to the same traffic stream.

Additional research done to test the sensitivity of the pavement analysis models to traffic load provides further insight into when the outcome of the pavement analysis is affected by changes in traffic loading. The following section discusses the development of the traffic loading scenarios used to perform those tests. The results of those pavement model sensitivity tests are provided in the final recommendations presented in Chapter 7 and include recommendations about when and where the LTPP SPS TPF load spectra can be used when site-specific load spectra are not available.

Recommended Scenarios for Sensitivity Tests of Pavement Design Models

Because it is practically impossible to test the sensitivity of the MEPDG to all possible combinations of site-specific vehicle streams and vehicle classification rule sets, a decision was made to identify and test a set of the worst cases that are likely to result in the maximum differences in traffic load observed when using one consistent classification rule set versus using a load spectra based on the LTPP WIM rule set and truck volume data collected using a State-specific vehicle classification rule set. The logic behind this approach is that if pavement design outcomes are not sensitive to these extreme differences, then all other less extreme differences can be considered insignificant. On the other hand, if significant differences are observed, more detailed site-specific analyses may be required.

To test the sensitivity of the pavement design models to differences in loading estimates resulting from application of different classification rules, five alternative traffic loading inputs were identified. The recommended loads are drawn directly from TPF sites. They include two load estimates under very heavy loading conditions and three load estimates under fairly light loading conditions.

The five input scenarios are intended to examine how significantly different the predicted pavement design and analysis outcomes are when different traffic loads are input under similar design conditions. The five recommended traffic loading scenarios are described below.

Heavy loading conditions—use a site (Arkansas SPS 2) with very heavy truck loads, using the LTPP spectra from that site combined with the following:

• State classification rule set that computes volumes that result in a very high traffic load estimate (California AVC).
• State classification rule set that computes volumes that result in a very low traffic load estimate Washington WIM).

Light loading conditions—use a site (Arizona SPS 1) with light truck loads, using the LTPP spectra from that site combined with the following:

• LTPP traffic volume (base case).
• State classification rule set that computes volumes that result in a very high traffic load estimate (California WIM).
• State classification rule set that computes volumes that result in a very low traffic load estimate Washington WIM).

Note that a baseline heavy volume condition is not recommended because the size of traffic loading errors for high traffic load conditions is very modest. Therefore, little knowledge is expected to be obtained from those additional pavement analysis runs.

PART II: SENSITIVITY OF PAVEMENT DESIGN MODELS TO DIFFERENCES IN VEHICLE CLASSIFICATION

Part II of this report explores and quantifies the sensitivity of the pavement design models to the differences in traffic load inputs associated with the application of the different vehicle classification rule sets. Specifically, the analyses describe the size and practical significance of differences in predicted pavement life and pavement depth caused by use of load spectra computed using the LTPP classification rule set instead of the State-specific rule set used to collect the truck volume data. These differences are then used to refine the recommendations in the previous chapter regarding when the SPS TPF load spectra can be used in pavement analyses without negatively affecting the accuracy of those analysis results.

Objectives

The goal of the analyses presented in this portion of the report is to explore and quantify the sensitivity of the pavement design models to the differences in traffic inputs associated with the application of the different vehicle classification rule sets.^(1,2) Specifically, the objective is to understand the impact on predicted pavement performance of potential errors in the traffic load inputs that result from combining vehicle classification data collected using a State-defined non-LTPP vehicle classification rule set with axle load spectra collected using the LTPP classification rule set.

The implication of differences between vehicle classification rule sets was evaluated from the perspective of practical impact on pavement design outcomes. Differences in pavement thickness (asphalt concrete (AC) or Portland cement concrete (PCC) surface layer) predictions of more than 0.5 inches were considered significant.

The findings from this study could be used to define areas of applicability of SPS TPF axle loading as surrogate or default axle loading data for General Pavement Studies (GPS) sites that do not have accurate axle loading information but have vehicle classification data collected using non-LTPP vehicle classification rule sets. Specifically, the following questions could be answered:

• Can LTPP GPS sites use WIM data from the SPS sites that use the LTPP classification rule set?
• Does it make any practical difference if the various class rule sets at the GPS sites differ from the LTPP classification rule set?

Organization of Part II

This portion of the final report contains three chapters. Chapter 5 examines the sensitivity of pavement design models to differences in traffic inputs developed using several vehicle classification rule sets identified at the end of chapter 4. These classification rule sets and specific LTPP sites were chosen because they produced the maximum difference in total loading between LTPP and other vehicle classification rule sets when applied at the same site.

Chapter 6 focuses specifically on the sensitivity of pavement design models to variations in predicted volumes of Classes 5 and 8 vehicles. The purpose of this analysis was to identify scenarios in which differences in Classes 5 and 8 vehicle counts, primarily owing to inclusion of additional lightweight vehicles, would matter for pavement design (i.e., produce significant difference in pavement design outcomes).

Finally, chapter 7 presents detailed conclusions based on the analyses findings provided at the end of chapters 5 and 6. Chapter 7 provides generalized conclusions with regard to the applicability and limitations of using loading data from the LTPP SPS TPF study for the sites that have vehicle classification and volume data collected using a non-LTPP classification rule set.

¹⁰ These percentage error estimates are computed as two times the standard deviation of the Class Ratio observed in tests performed for this project (see figure 15 through figure 17), assuming normal distribution of errors.

¹¹ A much larger source of error is likely to be the error associated with selecting load spectra from different site(s) and applying them to a site that may have different loading characteristics. This is the subject of a different LTPP analysis task. These errors are caused when non-site-specific load spectra are used, causing the differences in the percent of loaded trucks (or the percent of illegally loaded axles) between the actual site and what is present in the non-site-specific load spectra. This is likely to create far larger errors in the traffic load estimate than would be caused by the use of mismatched vehicle classification rule sets.