MEPDG Traffic Loading Defaults Derived From Traffic Pooled Fund Study

Chapter 9-MEPDG SENSITIVITY TO DIFFERENT AXLE LOAD SPECTRA CLUSTERS

BACKGROUND

Causes of Differences in Axle Load Spectra

Analysis of WIM data from the LTPP Program indicates that axle loading distributions vary between different axle group types, truck types (vehicle classes), and roadways. Differences in axle load spectra between different axle group types are attributed to the differences in axle configuration and different axle load limits that apply to single, tandem, tridem, and quad axles. Differences in the axle load spectra for the same axle group type between different truck types (vehicle classes) are attributed to differences in truck body structure and vehicle length (single unit versus multiple unit trucks) and the purpose of the axle (i.e., steering or load-carrying axle).

In addition, the commodities typically carried by different truck types affect axle weights (tandem axle weight on log hauler versus class 8 truck with short trailer). Bulky but lightly weighted commodities result in lighter axle loads than heavier commodities for the same type of axle and vehicle class (e.g., cereal boxes versus soda cans).

Even for the same vehicle class and axle group type, axle weights vary based on local conditions, such as the following:

• The differences in truck size and weight laws between States can result in different truck body/configurations for the same FHWA truck class.
  
  
• The presence and effectiveness of weight enforcement activities in the State and the percentage of through trucks versus local delivery trucks.
  
  
• The number of roadway lanes (outer lanes typically are heavier loaded than inner lanes on multilane roads).

It is also observed that these site conditions are more likely to have a significant effect on load spectra for locations that have a high percentage of intrastate traffic and a low percentage of interstate traffic.

Load Spectra Clusters

To test the sensitivity of the MEPDG outputs to traffic load, the project team developed a series of different NALS for use as inputs to the MEPDG program. These inputs were developed using axle weight collected at the 26 SPS TPF WIM sites.

The 26 sets of SPS TPF load spectra (i.e., load spectra for each class of vehicle and type of axle) were summarized into a limited number of NALS groups based on the similarities in their expected damaging effect on pavement deterioration. Expected damaging effect was determined using the RPPIF statistic discussed in chapter 8. Essentially, these groups represented different axle loading conditions (by type of axle and class of vehicle) that could be considered light to heavy within a given vehicle class and axle type, where light and heavy were determined based on the observed differences in RPPIF for each of the SPS TPF sites. The grouping (clustering) process used to create these NALS was similar to that described in chapter 8, except that the maximum allowed distance between mean RPPIF for nearest neighbor clusters (i.e., the break points between clusters) was determined using different criteria than for the spectra developed as new NALS defaults.

The initial criterion that the project team used was to make up to five clusters for each type of axle and class of vehicles while allowing for some flexibility in that number. To decide how to create the groups and how many groups to create, first a mean RPPIF was computed for each load spectra. The mean RPPIF was then used to find load bin for which the mean RPPIF andW factor were equal. If the difference between two mean RPPIF values for nearest neighbor load spectra groups was less than the difference between two consecutive W factor values, the load spectra groups were combined, even if this meant there were less than five groups. Similarly, for some heavy axle group types, if the mean RPPIF of individual sites was more than 1.0 RPPIF different than any group, these sites were left as independent groups even if that meant retaining more than five groups.

The end result was a diverse set of NALS groups from light to heavy which allowed a detailed analysis of sensitivity of the MEPDG to different traffic loading conditions. A list of the SPS TPF sites included in different load spectra clusters for this sensitivity test is included in appendix C of this report.

Vehicle classes and axles types (referred to as class-axle) where several statistically different axle loading conditions were identified are marked as "Y" in table 25.

*Based on LTPP vehicle classification scheme. Blank cells indicate data are unavailable.

Each of the 25 class-axle combinations had three to six clusters representing different axle loading conditions. The cluster that had the largest number of SPS TPF sites for each class-axle was called "typical." RANALS for the sites with the same cluster assignment were averaged to compute representative NALS for each cluster for each of the 25 class-axle combinations, resulting in 109 NALS used for the sensitivity analysis.

SENSITIVITY ANALYSIS OF LOAD SPECTRA CLUSTERS

Analysis Objective and Approach

The goal of the analysis was to assess if use of NALS representing different loading conditions, as determined based on statistical clustering analysis, would result in different MEPDG outcomes for a given class-axle combination. Significant differences in outcomes using different load spectra clusters for a given vehicle class and axle group type would support the need for alternatives to a single set of default load spectra, while the absence of differences would indicate that statistically identified load spectra clusters could be combined for a given vehicle class and axle group type.

The following criteria were used to identify significant differences between MEPDG design outcomes:

• Pavement thickness, as determined to support the selected design life at the specified level of reliability. A thickness difference for the top layer (AC or PCC) of over 0.5 inch was selected as significant for this analysis.
  
  
• Pavement life, as determined when the value in one of the performance criteria reaches its terminal value at the selected level of reliability for a fixed pavement structure. Pavement life differences of over 20 percent of design life were selected as significant for this analysis based engineering judgment.

All the pavement section designs were hypothetical. Where applicable, material types and design features were based on recommendations provided in the NCHRP 1-37A report and on typical values encountered in the LTPP database.⁽³⁾

This analysis was also used to provide insights into what class-axle combinations were most likely to have the most impact on pavement design when using the MEPDG. This was accomplished by comparing pavement thicknesses computed using typical axle load spectra with results predicted when load spectra was adjusted to a different load cluster default (e.g., keeping the axle load spectra for all vehicle classes and axle group types as typical in the base MEPDG run and then changing the typical spectra to the heavy spectra for class only 9 tandems in a subsequent MEPDG run).

MEPDG Traffic Inputs

Sensitivity of MEPDG outcomes to NALS was investigated independently for each vehicle class and axle group type. Considering that contribution of vehicle classes to total truck volume varies, the first step in selecting traffic inputs for analysis was to identify a most significant traffic volume by vehicle class scenario for each of the 25 class-axle combinations. The most significant scenario was defined to maximize the effect of a given vehicle class and axle group type on pavement performance. Truck volume by class data from all available LTPP sites (SPS and GPS combined) were used to define AADTT and the VCD for each truck class-axle group type combination using the following two criteria:

• Maximum percentage of trucks in a given class.
  
  
• Maximum AADTT volume.

The goal was to identify realistic truck traffic volume scenarios in the LTPP database that are likely to cause maximum damage for each specific class of truck and axle group type. Since realistic truck volume by class scenarios were used, some classes had much higher maximum volumes then others. For example, maximum class 9 volumes far exceeded maximum class 6 volumes in our tests because those were the conditions observed in the LTPP database. As a result, MEPDG sensitivity was influenced by observed volumes of vehicles in given vehicle classes. Such results cannot be used directly to evaluate if the class 9 tandem axle load spectrum is more damaging than the class 6 type tandem axle load spectrum, but this analysis is useful to evaluate if different class 6 load spectrum clusters are likely to produce different pavement design outcomes under realistic traffic conditions.

The search for truck volume data was done separately for flexible and for rigid pavements based on the observation that for the pavement sections included in the LTPP database, rigid pavement sections were more likely to have higher truck volumes than flexible pavement sections.

Table 26 shows the AADTT and percent trucks used for the analysis for each vehicle class. For example, MEPDG flexible pavement sensitivity analysis for class 9 load spectra was conducted for AADTT equal to 3,043 with 85.7 percent class 9 trucks. The corresponding LTPP section that provided the data for the analysis is shown in table 25.

For each class-axle combination, three to six different NALS representing different load cluster groups (GPs) were used in the MEPDG sensitivity analysis. For each GP, NALS were computed by averaging RANALS for the SPS TPF sites included in a given cluster. This was done separately for each vehicle class and each axle group. As a result, each of the 25 class-axle combination had several load spectra identified for sensitivity analysis. These spectra were coded as GP1, GP2, GP3, ...up to GP6. For a given class-axle combination, GP1 represented the lightest load spectrum, and the GP with the highest index represented the heaviest load spectrum (GP3 to GP6 based on the number of load spectra groups identified for a given class-axle combination).

Load spectra clusters identified as typical for each class-axle combination were used to define overall typical load spectra for the development of base MEPDG designs. The typical cluster was defined as the one that had the largest number of SPS TPF sites assigned to it. The assumption was made that this cluster likely would represent the most frequently observed traffic loading condition for a given class-axle combination in the LTPP database.

MEPDG Pavement Designs and Analysis Execution

To conduct MEPDG sensitivity analyses, pavement designs were developed for different traffic input scenarios. The climatic condition for the MEPDG analyses was selected as the one likely to cause the worst pavement performance. Previous studies in which the LTPP database was used suggested that the wet-freeze condition is the one most likely to cause more damage to the pavement. ^(17,18)

The pavement design life was set to 15 years for flexible pavements and 20 years for rigid-based on average service life computed for LTPP GPS flexible and rigid pavement sections (prior to first major rehabilitation activity). Traffic volume and class distribution were set according to table 26. MEPDG outputs were obtained using the 90 percent reliability option (default MEPDG option).

The design criteria used in the analysis are shown in table 27. The terminal values were based on defaults included in the MEPDG version 1.1 software. Failure modes were selected based on observed sensitivity of MEPDG models calibrated to global conditions. Different failure modes (critical distresses) and sensitivity analysis outcomes are possible for models calibrated to local conditions.

Exploratory MEPDG analysis involving joint faulting indicated that this distress is much more sensitive to the number of heavy repetitions (i.e., volume of heavy trucks) than to the difference in loading within a given heavy truck class. This is because pumping of base and subbase/subgrade material is a precursor of faulting, and that mechanism is highly sensitive to the number of load applications. In addition, the MEPDG provides detailed design guidelines for selection of erosion-resistant base material and dowel design to mitigate the development of this distress mode. Therefore, it is not expected that differences in the NALS, within a given truck class, would result in significantly different MEPDG design outcomes for properly designed PCC joints.

All design inputs selected were default values in the MEPDG software, with the exception of design features shown in table 28 for flexible pavement and table 29 for rigid pavement. Three designs were considered for flexible pavements. The reason for the use of these design inputs was to study MEPDG sensitivity under different failure modes: longitudinal top-down cracking (F1), rutting (F2), and bottom-up alligator cracking (F3). For rigid pavements, five designs were developed based on the MEPDG recommendations to minimize erosion and provide for load transfer at the joints based on observed AADTT levels.

F = Flexible.

R = Rigid.

MEPDG designs were developed for each case of truck class-axle combination and for each GP. This was done by changing the axle load spectra input and observing differences in pavement performance predictions or by adjusting thickness of the top structural pavement layer to achieve fixed pavement service life. For example, for the analysis of MEPDG sensitivity to class 9 tandem axle load spectra, the AADTT and VCD corresponding to class 9 in table 26 were used along with the typical load spectra for all vehicle classes and axle group types to develop a base design. Then, in subsequent MEPDG analyses, the typical load spectrum for class 9 tandems was changed so that the load spectra from different clusters (e.g., light or heavy) for class 9 tandems were used.

Since only the thickness of the surface layer was modified in each of these designs involving different load spectra clusters for a given vehicle type, the impact of different load spectra clusters on the design could be evaluated by simply comparing the differences in thickness of the surface layer.

Discussion of Findings

Flexible Pavements

The results for flexible pavements are provided in table 29 through table 31. Top-down cracking (F1 designs) was found to be the critical failure mode for all classes and combinations of axle load groups using globally calibrated MEPDG models with 90 percent design reliability. This finding is in part due to the high error term associated with top-down cracking that translates into a higher safety factor when designed for 90 percent reliability.

To study sensitivity of MEPDG bottom-up cracking mode to different axle load spectra clusters, a design that minimizes top-down cracking was developed: this is design F3 in table 28. However, even with this design, top-down cracking was significant. For the purpose of this investigation, top down cracking was not considered in the sensitivity analysis involving F3 designs and effect of NALS clusters on pavement failure in bottom-up failure mode was investigated for all classes and combinations of axle-load groups.

When the design inputs were adjusted to make pavements more susceptible to rutting as the failure mode (F2 designs), only those combinations of vehicle class and axle group type that produced a high percentage and volume of heavy trucks resulted in rutting failure. This applied to classes 9, 10, and 13. All others, despite the design inputs, ended up still failing in top-down cracking. Rutting is a primary distress in which a high percentage of heavy trucks have significant influence on magnitude and rate of deterioration. Consequently, only the class-axle combinations with high percentage of class 9 and above had rutting failure, when designed for highway speeds. For the roads with lower percentages of heavy truck volumes, rutting is possible under stop-and-go traffic conditions; however, these cases were not considered in this sensitivity analysis.

*High sensitivity of class 8 tandems is due to a single site, Florida SPS 1 site that forms tandem GP6. No other SPS TPF sites have this distribution.

Note: Bold values represent designs based on the typical NALS for all vehicle classes and axle group types. Blank cells indicate the group was not available.

*High sensitivity of class 8 tandems is due to a single site, Florida SPS 1 site that forms tandem GP6. No other SPS TPF sites have this distribution.

Note: Bold values represent designs based on the typical NALS for all vehicle classes and axle group types. Blank cells indicate the group was not available.

Note: Bold values represent designs based on the typical NALS for all vehicle classes and axle group types. Blank cells indicate the group was not available.

Under the truck volume and VCD conditions analyzed in this study, all classes except 5, 8, and 12 showed sensitivity to different axle load spectra that resulted in design thickness difference of 0.5 inch or more in one or more of the failure modes. Lack of class 5 load spectra sensitivity is explained by the low axle weight of class 5 vehicles, while low or no sensitivity of class 12 vehicles is explained by the low volumes and percentages of these vehicles compared to the other heavy vehicle types observed under typical traffic conditions. For class 8, only the Florida SPS-1 site showed design outcomes different from other clusters. This site had a high percentage of overweight class 8 trucks, and all other sites had much lighter axle weights.

When MEPDG sensitivity for specific axle group types was evaluated, NALS for single axles resulted in low MEPDG sensitivity for all vehicle classes, with the exception of class 8 (Florida SPS-1 site) and class 11 vehicles. Tandem axle loads showed the highest sensitivity among all axle group types. Pavement designs also were found to be sensitive to tridem and quad axles; however, for some cases, sensitivity was low primarily due to low percentages of tridem and quad axle load applications compared to other axles.

Class 9 tandems produced the greatest difference in pavement thickness prediction: up to 0.9 inch for bottom-up cracking, 0.8 inch for top-down cracking, and 2.6 inches in total rutting failure mode. This is the most frequently observed heavy truck class on U.S. interstates and principal arterial roads. Therefore, use of different axle load spectra for class 9 is likely to carry practical consequences in terms of costs and performance.

The summary of vehicle classes and axle group types that resulted in significant difference in design thickness and/or pavement service life based on MEPDG sensitivity analysis for flexible pavements is as follows:

• Class 4 tandem axles.
  
  
• Class 6 tandem axles.
  
  
• Class 7 tandem and tridem axles.
  
  
• Class 8 single and tandem axles (Florida site only).
  
  
• Class 9 tandem axles.
  
  
• Class 10 tandem and tridem axles.
  
  
• Class 11 single axles.
  
  
• Class 13 tandem, tridem, and quad axles.

The scenarios tested were based on truck traffic composition that had test class carrying the largest percentage of the total load observed among all LTPP sites. These scenarios represent the worst, rather than typical, condition that magnifies the effect of NALS associated with the test truck.

The results of the MEPDG sensitivity analyses for flexible pavements suggest that load spectra clusters defined in this study can yield differences in MEPDG-based designs that are of practical significance. Therefore, use of load spectra clusters that best represent axle loading condition at a given site is recommended for MEPDG-based flexible pavement designs.

Rigid Pavements

The results for rigid pavements are provided in table 33. Among all of the class-axle combinations investigated, only tandem axles in classes 4, 8, 9, and 13 had two or more cluster groups producing design slab thickness difference of 0.5 inch or more. Class 9 tandems produced the highest difference in pavement thickness prediction: 1.1 inches. Therefore, the selection of different load spectra for class 9 carries practical consequences in terms of rigid pavements costs and performance. Class 8 sensitivity was due to a single site, the Florida SPS-1 site. This site has very high percentage of overweight class 8 trucks. No other SPS TPF sites had this distribution.

From the service life perspective, observations were similar to the thickness analysis. In addition to the sensitive class-axle instances reported for thickness analysis, NALS clusters for class 6 tandem axles also showed a significant impact on service life, which was defined as differences in service life above 20 percent, or 4 years. These differences in design service life could have important economic impact on the life cycle cost of the pavement.

The summary of vehicle classes and axle group types that resulted in significant difference in design thickness and/or service life predictions based on MEPDG sensitivity analysis for rigid pavements is as follows:

• Class 4 tandem axles.
  
  
• Class 6 tandem axles.
  
  
• Class 8 tandem axles.
  
  
• Class 9 tandem axles.
  
  
• Class 13 tandem axles.

The results of MEPDG sensitivity analyses for rigid pavements suggest that load spectra clusters can yield differences in MEPDG-based designs that are of practical significance, specifically in the case of NALS for tandem axles. Since none of the single, tridem, or quad axle clusters resulted in significant changes in MEPDG outcomes, there is no evidence from this study that multiple axle loading defaults for single, tridem, or quad axles would result in significant benefits for rigid pavement designs. The scenarios tested were based on truck traffic composition that had test class carrying the largest percentage of the total load observed among all LTPP sites. These scenarios represent the worst, rather than typical, condition that magnifies the effect of NALS associated with the test truck.

*High sensitivity of class 8 tandems is due to a single site, Florida SPS-1that forms tandem GP6. No other SPS TPF sites have this distribution.

Note: Bold values represent designs based on typical NALS for all vehicle classes and axle group types. Blank cells indicate the group was not available.

Summary of Findings

Different MEPDG distress prediction models exhibit different relationships between design thickness and service life due to differences in pavement failure mechanics, as well as transfer functions used to relate mechanistic predictions to pavement distresses. In the case of rigid pavements, where the highest difference in the class 9 tandem axle scenario was about 1.1 inches, the impact in service life was over 14 years. Conversely, the same class-axle scenario for flexible pavement had the highest thickness difference of 0.8 and 2.6 inches for top-down cracking and rutting, respectively, but yielded differences in service life of 10 and 6 years, respectively. This finding emphasizes the importance of considering both design thickness and service life in the evaluation of the axle load spectra clusters.

Findings regarding individual axle group types and the impact of using different NALS clusters on MEPDG-based pavement design and performance prediction include the following:

• Single axles: The majority of analyses involving single-axle load spectra did not result in significant differences in pavement design thickness or service life. The exceptions are findings for bottom-up cracking in flexible pavements involving classes 8 (case of high overloads in Florida only) and 11. This finding does not support the need for multiple load spectra defaults (on the national level) for single axles, except for class 11.
  
  
• Tandem axles: Most of the tandem axle load spectra clusters resulted in significant differences in design thickness or service life of both flexible and rigid pavements. Exceptions are tandems for classes 5, 8, and 12 as follows:
  
  
  • Class 5 tandems are allowed by the LTPP vehicle classification scheme (light duty trailers on pick-up trucks or sport utility vehicles) and are expected to be light. These tandems are atypical for the majority of class 5 vehicles. Hence, they represent a very small percentage of axles and therefore cause insignificant damage.
    
    
  • Sensitivity of class 8 was found to be due to a single site located in a Florida agricultural region that had very different class 8 single and tandem NALS distribution compared to other SPS TPF sites, large percentage of overloaded tandem axles (40,000-44,000 lb). Use of alternative class 8 NALS defaults would not benefit MEPDG design accuracy for the sites that do not have significant percentages of overloaded class 8 tandems.
    
    
  • Class 12 has moderate-to-heavy loaded tandem axles, but for the VCD cases observed in LTPP database, the percentile contribution of class 12 trucks by volume was too small to produce significant variation in design thickness or service life. This finding may be different for states or regions that utilize high volumes of class 12 trucks.

This finding supports the need for multiple load spectra defaults for tandem axles for all vehicle classes except classes 5, 8, and 12.

• Tridem axles: All class-axle combinations that involved tridem load spectra clusters (classes 7, 10, and 13) resulted in differences in design thickness or service life for flexible but not for rigid pavements. This finding supports the need for multiple load spectra defaults for tridem axles to be used in flexible pavement design.
  
  
• Quad axles: With the exception of class 13 results for flexible pavements (rutting failure mode), no significant differences in pavement design outcomes were found when different quad axle load spectra clusters were used. This can be explained by the low percentages of quads, compared to other axle group types observed in the LTPP database, which is insignificant to produce large variation in design thickness or service life. This finding may be different for States or regions that utilize high volumes of trucks with quad axles. This current finding does not support the need for multiple load spectra defaults for quad axles, with the exception of class 13 quads when used for flexible pavement design. This conclusion is influenced by the vehicle classification data available at the time of study. However, in summer 2012, LTPP made a change to how vehicle classes 7, 10, and 13 are being classified, which may lead to a different conclusion in the future.

Findings regarding different vehicle classes and the impact of using different axle load spectra clusters on MEPDG-based pavement design and performance prediction are as follows:

• Classes not sensitive to NALS clusters: This finding does not support the need for multiple load spectra defaults for classes 5, 8, and 12.
  
  
  • Class 5 load spectra clusters for all axle group types did not result in significant differences in design thickness or service life in both pavement types. Two main reasons are: (1) class 5 trucks are light; the RPPIF for single and tandem clusters are among the lightest found in the all clusters evaluated in this study, and (2) although there were four single axle clusters and three tandem clusters identified through statistical clustering for this class, the differences among RPPIF's are very small, resulting in the low MEPDG sensitivity.
    
    
  • Class 8 load spectra clusters for single and tandem axle group types did not result in significant differences in design thickness or service life in both pavement types, with the exception of spectra for one site in Florida where many overloaded axles were observed. Class 8 axles observed at all other sites were lightly or moderately loaded.
    
    
  • Class 12 load spectra clusters for all axle group types did not result in significant differences in design thickness or service life in both pavement types. Although class 12 axles carry heavy loads, for the VCD cases observed in the LTPP database, the percentage contribution of class 12 was too small to warrant significant variation in design thickness or service life. This finding may be different for States or regions that utilize high volumes of class 11 or 12 trucks.
    
    
• Classes sensitive to NALS clusters: This finding indicates the need for multiple load spectra defaults for classes 4, 6, 7, 9, 10, 11, and 13.
  
  
  • MEPDG outcomes were sensitive to different NALS assigned to multi-axle groups (tandem, tridem, and quad) but not to single axle NALS groupings. Two exceptions where single axle NALS resulted in different outcomes are class 8 (case of high overloads only) and class 11 analysis for bottom-up cracking of flexible pavements.
    
    
  • Among all vehicle classes and axle group types, MEPDG outcomes were found to be the most sensitive to class 9 tandems. This is explained by high volumes of class 9 tandems coupled with variation of loads carried by class 9 vehicles.

Disclaimer

The conclusions presented in this chapter are based on the analysis of the traffic loading scenarios described in table 26. While the best attempt was made to select realistic traffic volume scenarios (AADTT and VCD) that would result in the highest sensitivity of pavement design outcomes to alternative load spectra groups, different traffic and pavement structure inputs may result in different analysis outcomes. It is recommended that State and local highway agencies conduct their own sensitivity analysis using State-specific truck traffic volume and classification inputs, as well as locally calibrated distress prediction models and local pavement design/performance criteria.

SENSITIVITY OF DIFFERENT ROAD AND PAVEMENT TYPES TO AXLE LOADING DISTRIBUTIONS

Analysis Objective

The purpose of this analysis was to determine if sensitivity of MEPDG outcomes to differences in NALS varies for pavements designed for different AADTT levels. The questions to be answered were as follows:

• Is pavement design sensitivity to NALS affected by pavement and road type?
  
  
• Would thin pavement structures designed for lower volume arterial roads show different sensitivity to NALS changes compared to thick pavement structures designed for high volume interstate roads?

Analysis Scope

The scope of the analysis was limited to MEPDG sensitivity to load spectra developed for class 9 tandem axles under different truck volume scenarios. This is by far the most dominant type of heavy axles observed on interstate roadways in the United States. Even for non-interstate highways, once class 5 axles are discounted due to their light weight, class 9 axles dominate the overall axle load spectra for most highway pavements. For the analyses conducted in this study, the presence of class 9 vehicles in the VCD was set to 85 percent to increase the potential MEPDG sensitivity to changes in loading of class 9 tandems. AADTT per lane values varied to capture the range of values observed on principal arterial interstates and rural highways.

Class 9 tandem load spectra representing typical, light, or heavy loading conditions developed based on SPS TPF WIM data were used in the analyses. For all other axle load spectra, the typical loading condition was kept constant through the analysis. Figure 37 shows class 9 tandem load spectra for the three identified conditions.

Figure 37. Graph. Class 9 tandem load spectra for identified clusters.

Several typical pavement designs were identified and analyzed for different AADTT levels. Both rigid and flexible pavement designs were included. With the exception of thickness of AC layer or PCC slab thickness, which varied by design, the design inputs shown in table 28 and table 29 were used in this MEPDG sensitivity study.

Analysis Execution

To conduct MEPDG sensitivity analyses, pavement designs were developed for different traffic input scenarios using the 90 percent reliability option (default MEPDG option). The wet-freeze climatic condition was used for all analyses as the condition most likely to cause more damage to the pavement based on previous LTPP studies.^(17,18) The pavement design life was set to 15 years for flexible pavements and 20 years for rigid based on average pavement design life observed for LTPP GPS sections. A range of AADTT per lane values was determined based on analysis of VCDs from LTPP database for the sites that had high percentage of heavy trucks, as presented in figure 38 through figure 40.

For each AADTT level, the pavement structure was designed twice. The first design was done using the light NALS for class 9 tandems and typical for all other class-axle load spectra. In the second design, the load spectrum for class 9 tandems was changed to heavy, while all other traffic inputs were kept the same. Light and heavy spectra were based on the lightest and the heaviest load spectra clusters for class 9 tandems, respectively. Only the thickness of the surface layer was modified in each of these designs. These two loading conditions resulted in the thinnest and thickest surface layer. The impact of different class 9 tandem load spectra on the design was evaluated by comparing the differences in the surface layer design thickness. The results of the analysis and relevant conclusions are provided in the following section.

Discussion of Findings

Flexible Pavements

The changes in design thicknesses using light and heavy NALS for class 9 tandems were investigated for a range of AADTT values and pavement failure modes (longitudinal top-down cracking, bottom-up alligator cracking, and rutting). Figure 38 illustrates the variations of design thickness versus AADTT for the top-down cracking failure mode. Figure 39 shows the variation of design thickness versus AADTT for the bottom up cracking failure mode. Figure 40 shows the variations of design thickness versus AADTT for the rutting failure mode.

Figure 38. Graph. Results of AC layer thickness sensitivity to class 9 load spectra for flexible pavements with top-down cracking failure mode.

Figure 39. Graph. Results of AC layer thickness sensitivity to class 9 load spectra for flexible pavements with bottom-up cracking failure mode.

Figure 40. Graph. Results of AC layer thickness sensitivity to class 9 load spectra for flexible pavements with rutting failure mode.

In the case of cracking failure modes, the difference in AC thickness between lightly and heavily loaded designs is somewhat uniform at about 1 inch. For practical purposes, the impact of class 9 load spectra was found significant for all AADTT levels investigated (i.e., difference in thickness higher or equal 0.5 inch).

In the case of rutting failure mode, it was only possible to investigate designs with AADTT values above 1,500. Below this level, the failure mode shifted from rutting to cracking. This is explained by globally calibrated models used and the design properties selected. The differences in thickness were much higher than previously seen for cracking, ranging from 1.9 to 2.7 inches for the AADTT analyzed. However, the approach of mitigating rutting by increasing AC thickness only may not be practical.

Overall, the results of flexible pavement analysis indicate that for all AADTT levels and design types investigated, the variation in load spectra between light and heavy loading conditions resulted in different AC surface thicknesses with practical consequences (i.e., differences higher than 0.5 inch). This indicates the importance of accurate axle loading characterization for AC pavements designed for all AADTT levels if heavy trucks dominate VCD.

Rigid Pavements

Similar to the flexible pavement analysis, the design thicknesses of PCC slab predicted based on the light and heavy load conditions of class 9 tandems were compared. Five different JPCP structural designs were used based on AADTT levels, as summarized in table 29. Figure 41 shows the PCC slab design thicknesses predictions for different AADTT levels.

There was only one critical distress observed throughout the analysis of all design cases: transverse slab cracking. As expected, the design thickness increased with the increase in traffic volume for both load conditions. However, the difference in thickness between the two loading conditions stayed almost the same for different AADTT levels (slightly higher difference for high AADTT compared to low AADTT). Even at the lowest truck volume level, when AADTT per lane was 250 vehicles and the slab thickness varied from 7.9 to 8.8 inches, the difference between the two load conditions was 0.9 inch. This difference is considered significant for practical purposes.

Figure 41. Graph. Results of PCC slab thickness sensitivity to class 9 load spectra for rigid pavements.

Conclusions

The results from the analyses for flexible and rigid pavements suggest that differences in design thickness will occur with practical implications for construction when one chooses a lightly loaded class 9 tandem load spectrum or a heavily loaded class 9 load spectrum, with no regards to truck traffic volume and pavement type, and pavement failure mode, provided the percentage of class 9 trucks remains high compared to other truck classes (class 5 excluded). For the range of traffic volumes considered in this analysis, differences in AC design thickness were observed between 2.7 and 0.8 inches depending on which load spectra and traffic level were chosen and which failure model governs the design. In the case of rigid pavements, the difference in design thickness was observed between 0.9 and 1.2 inches.

Based on the analysis findings, the following answers could be provided to the questions posted in the research objective:

• Is pavement design sensitivity to traffic loading affected by pavement and road type? All the designs showed sensitivity to difference in NALS for the class 9 trucks high enough to result in thickness differences that would have practical significance on the cost of the project. This indicates the importance of accurate characterization of axle loading for the heavy tuck types for all pavements types designed for all road types that have heavy trucks dominating VCD.
  
  
• Would thin pavement structures designed for lower volume arterial roads show different sensitivity to NALS changes compared to thick pavement structures designed for high volume interstate roads? Based on the differences in predicted thickness of the top structural layer (AC layer thickness or PCC slab thickness), there is no strong evidence that thin pavement structures designed for lower volume arterial roads are more or less sensitive to changes in NALS than thick pavement structures designed for high volume interstate roads, provided that each structure is adequately designed for a given truck volume and VCD. The MEPDG results indicate that change of load spectra results in similar changes of layer thickness for all AADTT levels, with pavement designed for lowest truck volumes showing slightly lesser differences in design thickness (in absolute values [inches], not in percentages). The magnitude of the difference in layer thickness is a function of pavement type and failure mode.

ANALYSIS OF TRUCK VOLUME AND DISTRIBUTION SCENARIOS LEADING TO HIGH SENSITIVITY OF MEPDG OUTCOMES TO CLASS 9 NALS

Analysis Objective and Scope

The purpose of this analysis was to identify traffic conditions (combinations of truck volume and VCD characteristics) when differences in load spectra defaults have a significant impact on MEPDG pavement design and analysis outcomes. The analysis was limited to tandem axles for class 9. These spectra were selected as the ones that have the most significant pavement damaging potential both due to the weight and the number of axle load applications. The question to be answered was, what combinations of AADTT and percentages of class 9 vehicles would require the use of axle load spectra that characterize a specific loading condition (versus use of global default) and what combinations of AADTT and percentages of class 9 vehicles would only marginally benefit from load spectra that accurately describe a specific loading condition?

These conditions would indicate at what point the presence of class 9 vehicles becomes too low to cause practical differences in pavement design outcomes with respect to differences in default axle load spectra selection.

Analysis Execution

This analysis started with the evaluation of truck volume data available in the LTPP database. Table 34 and table 35 show the distribution of LTPP sites by AADTT and percentage of class 9 trucks observed for LTPP flexible and rigid pavement sections. This information was used to identify plausible AADTT and class 9 percentage scenarios for the analysis. Based on LTPP information available with regard to AADTT, VCD, and pavement type, a total of 841 LTPP sections' AADTT and VCD values were analyzed from all the SPS and GPS sites. The computations were done based on the most recent year of data available in the LTPP SDR 24 database.

These AADTT ranges and class 9 percentages were used to identify truck volume scenarios. Once truck volume scenarios were identified, tandem axle load spectra clusters for class 9 vehicles that resulted in the thinnest (light NALS) and thickest (heavy NALS) pavement sections were selected for analysis, while typical NALS were used for all other vehicle classes and axles types (see figure 37).

AC and JPCP pavement structures were designed for each truck volume and NALS scenario, and differences in the resulting pavement thicknesses were analyzed. With the exception to thickness of AC layer or PCC slab thickness, which varied by design, the design inputs shown in table 28 and table 29 were used in this MEPDG sensitivity study.

Discussion of Findings

Flexible Pavements

Three design cases were used: one to mitigate rutting failure, one to mitigate top-down cracking failure, and one to mitigate bottom-up cracking failure. For each of these design cases, the analysis process began with the design scenario representing the highest AADTT and class 9 percentage identified in table 36 (AADTT per lane of 3,000 and 85 percent class 9 vehicles). Using the selected traffic volume scenario, the pavement was designed twice: once using load spectra with light class 9 tandem NALS and another time using heavy class 9 tandem NALS. After both designs were completed, the differences in surface layer thickness were computed and reported in table 36 (value = 0.8 inch). In subsequent steps, the AADTT and percentage class 9 were reduced in fixed steps. For each new AADTT class 9 scenario, pavement designs were adjusted to new traffic levels and the difference in top layer thickness (between light and heavy class 9 tandem NALS scenarios) was computed.

Table 36 provides the thickness difference matrix for the design case in which top-down cracking was observed as critical distress failure mode. The results suggest that the difference in design thickness between the designs using lightly and heavily loaded class 9 tandem spectrum remains at or above 0.5 inch at all levels for all the cases tested. This thickness was considered as the limit for a practical difference for this investigation.

Note: Blank cells indicate that no LTPP AC sections had this combination of AADTT and percent class 9.

Table 37 provides the thickness difference matrix for the design case in which bottom-up cracking was observed as critical distress failure mode. The results suggest that the difference in design thickness between the designs using lightly and heavily loaded class 9 tandem spectrum remained at or above 0.5 inch at all AADTT levels that had 45 percent or more on class 9 vehicles in VCD (for classes 4 through 13).

Note: Blank cells indicate no instances were recognized.

Table 38 provides the thickness difference matrix for the design with rutting failure mode. Similar to what was found in the cracking failure analysis, the results suggest that when the design is driven by rutting failure, the impact of selecting the load spectrum can have significant consequences on the outcome of the design (variation in thickness). Even at the lowest level of AADTT and percentage class 9 investigated, the difference in surface layer thickness was well above the 0.5-inch threshold. In addition to rutting failure, some cells also exhibited roughness and top-down cracking failure as a concomitant. This happened due to design characteristics and because the roughness model in the MEPDG had components that were output from other distress models. When the volume of class 9 starts to diminish, and consequently the design thickness decreases, distresses such as cracking start to increase, which contributes to the increase in roughness observed in the predictions. Lower AADTT values were not tested for rutting failure mode, as this mode is dominant for roads with high volumes of trucks operated at highway speeds.

*These designs also failed in top-down cracking mode.

Note: Blank cells indicate that no LTPP AC sections had this combination of AADTT and percent class 9.

The results for flexible pavement designs indicate that even when lower a percentage of class 9 was considered in VCD and a lower volume of trucks was used, the variation in load spectra between lightly and heavily loaded resulted in different AC surface thicknesses with practical consequences (i.e., differences higher or equal to 0.5 inch). Moreover, if the design was governed by rutting, the differences in thickness were even more significant. One exception was for roads with less than 45 percent of class 9 vehicles that failed in bottom-up cracking mode.

Rigid Pavements

In the case of rigid pavements, only one failure mode was observed-slab cracking. The designs were adjusted based on the MEPDG recommendations (e.g., dowel bar diameters and spacing, base layer thickness, and material type) to mitigate the potential of joint faulting. The rigid pavement design characteristics were the same as the ones described in table 29.

Table 39 provides the differences in the predicted PCC pavement design thickness obtained from analysis of light and heavy class 9 tandem load spectra clusters using various AADTT and percentage of class 9 values. The results show that rigid pavements designs are sensitive to variation in NALS for class 9 tandem axles at all AADTT levels tested, especially when 15 percent or more class 9 vehicles are present in the VCD. Differences at or above 0.5 inch observed in table 39 were considered of practical significance.

Conclusions

MEPDG sensitivity to different NALS for class 9 tandem axle loads was investigated for various truck volume and class 9 percentage scenarios. The results for both rigid and flexible pavements suggest that the impact of selecting a lightly or heavily loaded spectrum for class 9 tandem axles is significant under most traffic scenarios included in the study. Overall, the results indicate that even when a low percentage of class 9 vehicles (e.g., 15 percent) is considered in VCD and a low volume of trucks is used (AADTT per lane is 250), the variation in load spectra between lightly and heavily loaded class 9 tandems resulted in significant differences in the AC surface thickness and PCC slab thickness.

The results of the analysis indicate that it is important to have accurate characterization of class 9 tandem axle load spectra under all traffic conditions observed for interstate and primary arterial roads, and that local knowledge of loading conditions is important for accurate pavement design.

ANALYSIS OF MEPDG OUTCOMES USING CLASS 9 NALS FOR 26 SPS TPF SITES

Analysis Objective and Scope

In this analysis, class 9 tandem NALS computed for 26 SPS TPF sites were used to predict pavement design life for several hypothetical pavement designs. The purpose of the analysis was to see if MEPDG outcomes would form clusters based on the class 9 tandem NALS inputs and whether outcomes obtained using NALS for the sites located on interstate roads would be different from NALS for the sites located on the non-interstate principal arterial roads. In addition, correlation RPPIF and percent heavy axles statistics and pavement design life was investigated.

The design inputs shown in table 28 and table 29 for high-volume roads (AADTT > 3,000, TTC1) were used in this MEPDG sensitivity study. Class 9 tandem NALS was the only parameter changing within a given MEPDG design.

Analysis Findings

Figure 42 through figure 45 show the relationship between MEPDG-predicted pavement life and one of the two traffic loading statistics: percentage of heavy loads or average RPPIF associated with a given class 9 tandem NALS. Each symbol on the plot represents one of 26 cases of class 9 tandem NALS used in MEPDG pavement life prediction. To investigate whether pavement life predictions group in clusters based on the road functional class, two separate symbols were used, one for interstates and another for non-interstate roads. In addition, percentage of heavy loads (over 80 percent of the legal limit) and average RPPIF values were used to see if these values would form clearly defined clusters or trends with respect to pavement life.

Figure 42. Graph. Results of pavement life prediction for bottom-up cracking mode.

Figure 43. Graph. Results of pavement life prediction for top-down cracking mode.

Figure 44. Graph. Results of pavement life prediction for rutting failure mode.

Figure 45. Graph. Results of pavement life prediction for slab cracking mode.

The results shown on the plots led to the following conclusions:

• Class 9 tandem NALS computed for RI and ROPA show similar ranges of heavy axle loads and similar ranges of pavement deign life predictions. Difference in average pavement design life using RI and ROPA NALS was less than 2 years. On average, ROPA NALS had higher percentage of light loads and loads over Federal legal limit, but these differences did not translate in significant differences in MEPDG outcomes. Therefore, no clustering of the MEPDG results based on road functional class were observed for these two road types.
  
  
• There is a strong corelation between RPPIF and pavement deign life-high RPPIF relates to short pavement design life, and low RPPIF relates to long pavement design life. RPPIF is therefore considered an effective statistic for grouping NALS that are likely to result in similar MEPDG outcomes (NALS with similar RPPIF values should be grouped togetther). Least squared difference statistic for linear regression models relating RPPIF to predicted pavement design life ranged from 94 to 96 percent for different distress types.
  
  
• A correlation was also observed between the percentage of heavy loads and the pavement deign life. Therefore, percentages of heavy loads could be used to assign site-specific NALS to the appropriate GP or to describe RPPIF-based NALS clusters.
  
  
• Results of pavement design life predictions based on different class 9 tandem NALS obtained from 26 SPS PFS sites showed fairly continuous outcomes without clearly defined clusters, with the exception of rutting, thus making it challenging to provide logical breakouts for load clusters to define light, moderate, and heavy loading conditions. In other words, changes in class 9 tandem spectra (and resulting MEPDG outcomes) from one SPS TPF site to another are gradual, without clear definition of groups or clusters of sites with similar outcomes.
  
  
• For the cases tested, a difference of 0.9 to 0.16 in the RPPIF statistic for class 9 tandem NALS resulted in a 20 percent difference in pavement life prediction. Pavement life prediction was most sensitive for the JPCP slab cracking mode, with a difference of 0.09 in the RPPIF statistic corresponding to a 20 percent difference in pavement life prediction. Total rutting was least sensitive to a difference in the RPPIF statistic for class 9 tandem NALS, with a 0.16 difference in the RPPIF statistic corresponding to a 20 percent difference in pavement life prediction.

RECOMMENDATIONS

MEPDG analysis of NALS developed for different vehicle classes and axle group types was conducted to determine what classes and axle load distributions cause the most impact on MEPDG outcomes. Specifically, differences in axle load distributions observed within each vehicle class and axle group type and the effect of these differences on MEPDG outcomes were evaluated. Based on the findings from the sensitivity analyses, the following recommendations are made with respect to developing default NALS based on LTPP SPS TPF data:

• Single axles: None of the single-axle load spectra clusters resulted in significant differences in design thickness or service life, with the exception of class 11. Therefore, the development of multiple load spectra defaults for single axles is not recommended, except for class 11.
  
  
• Tandem axles: Tandem-axle load spectra clusters for all vehicle classes, except classes 5, 8, and 12, resulted in significant differences in design thickness or service life predictions for both flexible and rigid pavements. Therefore, it is recommended that multiple load spectra default for tandem axles are to be developed and used in MEPDG applications except for classes 5, 8, and 12. Class 12 has heavy tandem axles and differences in load spectra that could result in differences in pavement thickness, for States or regions that utilize high volumes of class 12 trucks. In this case, multiple loading defaults for class 12 vehicles would be recommended.
  
  
• Tridem axles: All class-axle combinations that involved tridem load spectra clusters (classes 7, 10, and 13) resulted in differences in design thickness or service life for flexible but not for rigid pavements. Therefore, it is recommended that multiple default load spectra for tridem axles be developed and used in MEPDG applications involving flexible pavement design.
  
  
• Quad axles: With the exception of class 13 results for flexible pavements (rutting failure mode), no significant differences in pavement design outcomes were found when different quad axle load spectra clusters were used, mainly due to the low percentage of quads axles, compared to other axle group types. Therefore, the development of multiple load spectra defaults for quad axles is not recommended, except for class 13 quads when used for flexible pavement design. This finding may be different for States or regions that have high volumes of quad axles. In this case, multiple loading defaults for quad axles may be needed.
  
  
• Highest MEPDG sensitivity: Among all vehicle classes and axle group types, MEPDG outcomes were found to be the most sensitive to class 9 tandems. This is explained by high volumes of class 9 tandems compared to other class-axle combinations, coupled with high variation of loads carried by class 9 vehicles. Even when a low percentage of class 9 vehicles (15 percent) and a low volume of trucks (AADTT per lane is 250) were used in MEPDG analyses, the variation in load spectra between lightly and heavily loaded class 9 tandems resulted in significant differences in the AC surface thickness and PCC slab thickness. As little as 10 percent difference in the total percentage of heavy tandem axles could result in up to 20 percent difference in pavement design life for rigid pavements and 15 percent difference for flexible pavements. Therefore, it is recommended that pavement designers place high importance on accurate characterization of class 9 tandem axle load spectra under all traffic conditions observed for interstate and primary arterial roads when conducting MEPDG analysis.
  
  
• Low volume of certain truck classes: For specific pavement designs, when individual vehicle classes have very low volumes compared to other heavy truck classes, the choice of load spectra for those specific low-volume vehicle classes will not have an impact on pavement thickness. Conditions when different load spectra may have an effect on pavement design outcomes, as found through MEPDG sensitivity analysis, are summarized in table 23.
  
  
• NALS representing "special case" loading: In many cases, NALS that led to the most different MEPDG outcomes (thinnest or thickest pavement designs) were representing loading conditions observed in a single State or only at a specific SPS TPF site. While these NALS were very useful for MEPDG sensitivity analysis, it is recommended that these individual NALS or clusters be clearly identified and not recommended for general use as defaults, unless knowledge of local conditions supports use of these "special case" NALS.
  
  
• NALS clusters: Both RPPIF and percent heavy loads statistics were found to have high correlation with predicted pavement design life based on MEPDG method. These statistical parameters can be used to define/describe NALS clusters representing different axle loading conditions. It is recommended that any statistically defined clusters be further enhanced or revised based on the findings and recommendations from the MEPDG sensitivity analysis. Specifically, break down points similar to ones summarized in Table 23, as defined by difference between mean RPPIF values for two clusters, should be used in clustering analysis to identify clusters that are likely to result in significantly different MEPDG outcomes.

DISCLAIMER

All MEPDG sensitivity analyses were developed using the MEPDG version 1.1 software. The terminal values used for the design criteria were based on defaults included in this software. Failure modes were selected based on observed sensitivity of globally calibrated MEPDG models. Different failure modes (critical distresses) and sensitivities are possible for models calibrated to local conditions or for cases where different terminal values are used.