National Academies Press: OpenBook
« Previous: Analysis
Page 95
Suggested Citation:"Validation of Analysis Methods." National Academies of Sciences, Engineering, and Medicine. 2021. Investigating the Relationship of As-Constructed Asphalt Pavement Air Voids to Pavement Performance. Washington, DC: The National Academies Press. doi: 10.17226/26219.
×
Page 95
Page 96
Suggested Citation:"Validation of Analysis Methods." National Academies of Sciences, Engineering, and Medicine. 2021. Investigating the Relationship of As-Constructed Asphalt Pavement Air Voids to Pavement Performance. Washington, DC: The National Academies Press. doi: 10.17226/26219.
×
Page 96
Page 97
Suggested Citation:"Validation of Analysis Methods." National Academies of Sciences, Engineering, and Medicine. 2021. Investigating the Relationship of As-Constructed Asphalt Pavement Air Voids to Pavement Performance. Washington, DC: The National Academies Press. doi: 10.17226/26219.
×
Page 97
Page 98
Suggested Citation:"Validation of Analysis Methods." National Academies of Sciences, Engineering, and Medicine. 2021. Investigating the Relationship of As-Constructed Asphalt Pavement Air Voids to Pavement Performance. Washington, DC: The National Academies Press. doi: 10.17226/26219.
×
Page 98
Page 99
Suggested Citation:"Validation of Analysis Methods." National Academies of Sciences, Engineering, and Medicine. 2021. Investigating the Relationship of As-Constructed Asphalt Pavement Air Voids to Pavement Performance. Washington, DC: The National Academies Press. doi: 10.17226/26219.
×
Page 99
Page 100
Suggested Citation:"Validation of Analysis Methods." National Academies of Sciences, Engineering, and Medicine. 2021. Investigating the Relationship of As-Constructed Asphalt Pavement Air Voids to Pavement Performance. Washington, DC: The National Academies Press. doi: 10.17226/26219.
×
Page 100
Page 101
Suggested Citation:"Validation of Analysis Methods." National Academies of Sciences, Engineering, and Medicine. 2021. Investigating the Relationship of As-Constructed Asphalt Pavement Air Voids to Pavement Performance. Washington, DC: The National Academies Press. doi: 10.17226/26219.
×
Page 101
Page 102
Suggested Citation:"Validation of Analysis Methods." National Academies of Sciences, Engineering, and Medicine. 2021. Investigating the Relationship of As-Constructed Asphalt Pavement Air Voids to Pavement Performance. Washington, DC: The National Academies Press. doi: 10.17226/26219.
×
Page 102
Page 103
Suggested Citation:"Validation of Analysis Methods." National Academies of Sciences, Engineering, and Medicine. 2021. Investigating the Relationship of As-Constructed Asphalt Pavement Air Voids to Pavement Performance. Washington, DC: The National Academies Press. doi: 10.17226/26219.
×
Page 103
Page 104
Suggested Citation:"Validation of Analysis Methods." National Academies of Sciences, Engineering, and Medicine. 2021. Investigating the Relationship of As-Constructed Asphalt Pavement Air Voids to Pavement Performance. Washington, DC: The National Academies Press. doi: 10.17226/26219.
×
Page 104
Page 105
Suggested Citation:"Validation of Analysis Methods." National Academies of Sciences, Engineering, and Medicine. 2021. Investigating the Relationship of As-Constructed Asphalt Pavement Air Voids to Pavement Performance. Washington, DC: The National Academies Press. doi: 10.17226/26219.
×
Page 105
Page 106
Suggested Citation:"Validation of Analysis Methods." National Academies of Sciences, Engineering, and Medicine. 2021. Investigating the Relationship of As-Constructed Asphalt Pavement Air Voids to Pavement Performance. Washington, DC: The National Academies Press. doi: 10.17226/26219.
×
Page 106
Page 107
Suggested Citation:"Validation of Analysis Methods." National Academies of Sciences, Engineering, and Medicine. 2021. Investigating the Relationship of As-Constructed Asphalt Pavement Air Voids to Pavement Performance. Washington, DC: The National Academies Press. doi: 10.17226/26219.
×
Page 107
Page 108
Suggested Citation:"Validation of Analysis Methods." National Academies of Sciences, Engineering, and Medicine. 2021. Investigating the Relationship of As-Constructed Asphalt Pavement Air Voids to Pavement Performance. Washington, DC: The National Academies Press. doi: 10.17226/26219.
×
Page 108
Page 109
Suggested Citation:"Validation of Analysis Methods." National Academies of Sciences, Engineering, and Medicine. 2021. Investigating the Relationship of As-Constructed Asphalt Pavement Air Voids to Pavement Performance. Washington, DC: The National Academies Press. doi: 10.17226/26219.
×
Page 109
Page 110
Suggested Citation:"Validation of Analysis Methods." National Academies of Sciences, Engineering, and Medicine. 2021. Investigating the Relationship of As-Constructed Asphalt Pavement Air Voids to Pavement Performance. Washington, DC: The National Academies Press. doi: 10.17226/26219.
×
Page 110
Page 111
Suggested Citation:"Validation of Analysis Methods." National Academies of Sciences, Engineering, and Medicine. 2021. Investigating the Relationship of As-Constructed Asphalt Pavement Air Voids to Pavement Performance. Washington, DC: The National Academies Press. doi: 10.17226/26219.
×
Page 111
Page 112
Suggested Citation:"Validation of Analysis Methods." National Academies of Sciences, Engineering, and Medicine. 2021. Investigating the Relationship of As-Constructed Asphalt Pavement Air Voids to Pavement Performance. Washington, DC: The National Academies Press. doi: 10.17226/26219.
×
Page 112
Page 113
Suggested Citation:"Validation of Analysis Methods." National Academies of Sciences, Engineering, and Medicine. 2021. Investigating the Relationship of As-Constructed Asphalt Pavement Air Voids to Pavement Performance. Washington, DC: The National Academies Press. doi: 10.17226/26219.
×
Page 113
Page 114
Suggested Citation:"Validation of Analysis Methods." National Academies of Sciences, Engineering, and Medicine. 2021. Investigating the Relationship of As-Constructed Asphalt Pavement Air Voids to Pavement Performance. Washington, DC: The National Academies Press. doi: 10.17226/26219.
×
Page 114
Page 115
Suggested Citation:"Validation of Analysis Methods." National Academies of Sciences, Engineering, and Medicine. 2021. Investigating the Relationship of As-Constructed Asphalt Pavement Air Voids to Pavement Performance. Washington, DC: The National Academies Press. doi: 10.17226/26219.
×
Page 115
Page 116
Suggested Citation:"Validation of Analysis Methods." National Academies of Sciences, Engineering, and Medicine. 2021. Investigating the Relationship of As-Constructed Asphalt Pavement Air Voids to Pavement Performance. Washington, DC: The National Academies Press. doi: 10.17226/26219.
×
Page 116
Page 117
Suggested Citation:"Validation of Analysis Methods." National Academies of Sciences, Engineering, and Medicine. 2021. Investigating the Relationship of As-Constructed Asphalt Pavement Air Voids to Pavement Performance. Washington, DC: The National Academies Press. doi: 10.17226/26219.
×
Page 117
Page 118
Suggested Citation:"Validation of Analysis Methods." National Academies of Sciences, Engineering, and Medicine. 2021. Investigating the Relationship of As-Constructed Asphalt Pavement Air Voids to Pavement Performance. Washington, DC: The National Academies Press. doi: 10.17226/26219.
×
Page 118
Page 119
Suggested Citation:"Validation of Analysis Methods." National Academies of Sciences, Engineering, and Medicine. 2021. Investigating the Relationship of As-Constructed Asphalt Pavement Air Voids to Pavement Performance. Washington, DC: The National Academies Press. doi: 10.17226/26219.
×
Page 119
Page 120
Suggested Citation:"Validation of Analysis Methods." National Academies of Sciences, Engineering, and Medicine. 2021. Investigating the Relationship of As-Constructed Asphalt Pavement Air Voids to Pavement Performance. Washington, DC: The National Academies Press. doi: 10.17226/26219.
×
Page 120
Page 121
Suggested Citation:"Validation of Analysis Methods." National Academies of Sciences, Engineering, and Medicine. 2021. Investigating the Relationship of As-Constructed Asphalt Pavement Air Voids to Pavement Performance. Washington, DC: The National Academies Press. doi: 10.17226/26219.
×
Page 121
Page 122
Suggested Citation:"Validation of Analysis Methods." National Academies of Sciences, Engineering, and Medicine. 2021. Investigating the Relationship of As-Constructed Asphalt Pavement Air Voids to Pavement Performance. Washington, DC: The National Academies Press. doi: 10.17226/26219.
×
Page 122
Page 123
Suggested Citation:"Validation of Analysis Methods." National Academies of Sciences, Engineering, and Medicine. 2021. Investigating the Relationship of As-Constructed Asphalt Pavement Air Voids to Pavement Performance. Washington, DC: The National Academies Press. doi: 10.17226/26219.
×
Page 123
Page 124
Suggested Citation:"Validation of Analysis Methods." National Academies of Sciences, Engineering, and Medicine. 2021. Investigating the Relationship of As-Constructed Asphalt Pavement Air Voids to Pavement Performance. Washington, DC: The National Academies Press. doi: 10.17226/26219.
×
Page 124
Page 125
Suggested Citation:"Validation of Analysis Methods." National Academies of Sciences, Engineering, and Medicine. 2021. Investigating the Relationship of As-Constructed Asphalt Pavement Air Voids to Pavement Performance. Washington, DC: The National Academies Press. doi: 10.17226/26219.
×
Page 125
Page 126
Suggested Citation:"Validation of Analysis Methods." National Academies of Sciences, Engineering, and Medicine. 2021. Investigating the Relationship of As-Constructed Asphalt Pavement Air Voids to Pavement Performance. Washington, DC: The National Academies Press. doi: 10.17226/26219.
×
Page 126
Page 127
Suggested Citation:"Validation of Analysis Methods." National Academies of Sciences, Engineering, and Medicine. 2021. Investigating the Relationship of As-Constructed Asphalt Pavement Air Voids to Pavement Performance. Washington, DC: The National Academies Press. doi: 10.17226/26219.
×
Page 127
Page 128
Suggested Citation:"Validation of Analysis Methods." National Academies of Sciences, Engineering, and Medicine. 2021. Investigating the Relationship of As-Constructed Asphalt Pavement Air Voids to Pavement Performance. Washington, DC: The National Academies Press. doi: 10.17226/26219.
×
Page 128
Page 129
Suggested Citation:"Validation of Analysis Methods." National Academies of Sciences, Engineering, and Medicine. 2021. Investigating the Relationship of As-Constructed Asphalt Pavement Air Voids to Pavement Performance. Washington, DC: The National Academies Press. doi: 10.17226/26219.
×
Page 129
Page 130
Suggested Citation:"Validation of Analysis Methods." National Academies of Sciences, Engineering, and Medicine. 2021. Investigating the Relationship of As-Constructed Asphalt Pavement Air Voids to Pavement Performance. Washington, DC: The National Academies Press. doi: 10.17226/26219.
×
Page 130
Page 131
Suggested Citation:"Validation of Analysis Methods." National Academies of Sciences, Engineering, and Medicine. 2021. Investigating the Relationship of As-Constructed Asphalt Pavement Air Voids to Pavement Performance. Washington, DC: The National Academies Press. doi: 10.17226/26219.
×
Page 131

Below is the uncorrected machine-read text of this chapter, intended to provide our own search engines and external engines with highly rich, chapter-representative searchable text of each book. Because it is UNCORRECTED material, please consider the following text as a useful but insufficient proxy for the authoritative book pages.

95 Figure 2-52. Example of actual rutting performance curves of LTPP pavement sections. Validation of Analysis Methods Validation of Analysis Method 1 Using Pavement ME Design The strength of Analysis Method 1 was assessed using AASHTOWare Pavement ME Design software Version 2.5.5 (PaveME). Two subgroups of LTPP sections with varying climate, traffic, and pavement structure were selected to compare the study’s scatter plots trends to PaveME predicted performance trends over a range of as-constructed AV. Nine LTPP sections from five states were selected from the results of Analysis Method 1 and represented one subgroup with the expected performance response (as AV increase, distress increases) and the other with no influence on performance (AV have no significant effect on the distresses). Table 2-32 summarizes the sections that were selected for this assessment with their corresponding traffic level in million ESALs for a ten-year period. Traffic level for the sections under evaluation varies from 0.5 million ESALs to 2.2 million ESALs. The PaveME validation focused on comparing rutting and fatigue cracking responses. The PaveME predicted performance for transverse (thermal) cracking and for ride were easily identified as unreasonable values, so those performance characteristics were not included in the PaveME validation. Table 2-32. LTPP sections selected for PaveME software assessment. State Code Section ID Location Analysis Method 1 Subgroup 10-Year Cumulative Traffic (MESALs) 4 115 Mohave, AZ Dry-No Freeze, Low Traffic, Medium Pavement 1.8 4 116 Mohave, AZ Dry-No Freeze, Low Traffic, Medium Pavement 1.8 40 118 Comanche, OK Dry-No Freeze, Low Traffic, Medium Pavement 0.5 40 122 Comanche, OK Dry-No Freeze, Low Traffic, Medium Pavement 0.5 5 114 Poinsett, AR Wet-No Freeze, Low Traffic, Medium Pavement 2.2 5 117 Poinsett, AR Wet-No Freeze, Low Traffic, Medium Pavement 2.2

96 State Code Section ID Location Analysis Method 1 Subgroup 10-Year Cumulative Traffic (MESALs) 5 118 Poinsett, AR Wet-No Freeze, Low Traffic, Medium Pavement 2.2 22 116 Calcasieu, LA Wet-No Freeze, Low Traffic, Medium Pavement 0.5 51 118 Pittsylvania, VA Wet-No Freeze, Low Traffic, Medium Pavement 1.8 The methodology that was followed to generate PaveME input files was as follows. Pavement Structure and Material Properties Using the LTPP database, pavement structures in terms of layer type and thickness were modeled using the PaveME input criteria. PaveME using Level 1 input requires a complete laboratory characterization of all the pavement structure materials, i.e., AC, AC binder, base, and subgrade. The LTPP database contains partial information required for Level 1 input for some of the sections, but it was not complete enough to generate Level 1 input for all sections. For this validation, the simulations were constructed using Level 3 input data. Traffic and Climate PaveME requires traffic input that includes AADTT and distribution of different classes of vehicles. Since traffic information for the LTPP sections in the study only extracted available annual ESALs, the approach for this validation task input 100 percent Class 5 trucks and zero percent for all other truck classes, and 100 percent 18,000 lb. single axle load and zero percent for all other single axle loads. Using this approach, the cumulative heavy trucks generated as input in the PaveME software can be easily converted to ESALs. All the climate input data needed by the PaveME Software are available from weather stations that can be selected by providing the longitude and latitude of the projects (available in the LTPP InfoPave website) and selecting the closest stations. PaveME Simulations Once the LTPP pavement sections were modeled, the simulations were executed to obtain the predicted pavement performance in terms of pavement distresses. In addition, two additional simulations were executed for each LTPP section by varying the top asphalt layer as-constructed AV by -2% and +2%. Comparison of PaveME Predictions to Analysis Method 1 For Analysis Method 1, the rutting performance indicator was measured in inches after four years of traffic on the asphalt surface. The fatigue cracking performance indicator was percent cracking in the wheel path after 10 years of traffic. These two performance indicators for each section were compared to PaveME predicted pavement performance. Actual as-constructed AV for the sections under evaluation range from 2.1% to 10.3%. Table 2-33 shows a comparison of predicted rutting and fatigue cracking of the pavement sections obtained from the PaveME analysis and the measured Analysis Method 1 performance values and trends. From this table, it can be observed that the actual rutting data for all sections are less than 0.23 in (LTPP Section 40-122). The maximum fatigue cracking measurement is 45.4% (LTPP section 51-118).

97 The primary focus of the comparison is the trend in performance relative to the change in as-constructed AV. Table 2-33. Comparison of PaveME predicted trends to Analysis Method 1 observed trends. Section Subgroup (New Pvmt) Data Source As-constructed Air Voids Distress Rutting @ 4 Years (in) Cracking @ 10 Years (% lane area) 4-115 2-Low-Med PaveME -2% AV 0.20 1.5 Actual AV=8.7% 0.22 1.5 +2% AV 0.25 1.9 LTPP Analysis 1 0.10 met expectation 20.4 met expectation 4-116 2-Low-Med PaveME -2% AV 0.22 1.5 Actual AV=10.3% 0.25 1.8 +2%AV 0.28 3.0 LTPP Analysis 1 0.22 met expectation 32.4 met expectation 40-118 2-Low-Med PaveME -2% AV 0.20 1.6 Actual AV=7.5% 0.20 1.6 +2% AV 0.20 1.6 LTPP Analysis 1 0.20 met expectation 40.3 met expectation 40-122 2-Low-Med PaveME -2% AV 0.20 1.4 Actual AV=6.5% 0.20 1.4 +2% AV 0.22 1.4 LTPP Analysis 1 0.23 met expectation 30.6 met expectation 5-114 4-Low-Med PaveME -2% AV 0.30 18.0 Actual AV=7.6% 0.32 20.0 +2% AV 0.32 22.0 LTPP Analysis 1 0.20 No influence 44.7 Met expectation 5-117 4-Low-Med PaveME -2% AV 0.18 1.4 Actual AV=6.1% 0.18 1.4 +2% AV 0.35 50.0 LTPP Analysis 1 0.20 No influence 30.1 Met expectation 5-118 4-Low-Med PaveME -2% AV 0.35 2.0 Actual AV=8.5% 0.35 2.0 +2% AV 0.37 2.0 LTPP Analysis 1 0.21 No influence 23.4 Met expectation 22-116 4-Low-Med PaveME -2% AV1 NA NA Actual AV=2.1% 0.18 1.4 +2% AV 0.16 1.4

98 Section Subgroup (New Pvmt) Data Source As-constructed Air Voids Distress Rutting @ 4 Years (in) Cracking @ 10 Years (% lane area) LTPP Analysis 1 0.20 No influence 0.0 Met expectation 51-118 4-Low-Med PaveME -2% AV 0.30 1.9 Actual AV=9.9% 0.30 2.0 +2% AV 0.35 2.0 LTPP Analysis 1 0.20 No influence 45.4 Met expectation legend met expectation no influence Note 1: Since actual AV=2.1%, no PaveME analysis was conducted using -2% AV. In general, for the nine sections evaluated, PaveME prediction trends compare reasonably well to Analysis Method 1 trends based on the following observations. • PaveME rutting predictions are very similar to Analysis Method 1 results for five out of nine sections. For sections 4-115 and 5-118, PaveME over-predicts rutting, and for section 5-114 and 51- 118 PaveME under-predicts rutting. • PaveME rutting predictions at different as-constructed AV show the expected trend. As AV increase, rutting increases, with the exception of section 40-118, which shows no change in rutting at different AV. • PaveME under-predicts fatigue cracking values for all of the sections with the exception of section 22-116. Analysis Method 1 measured fatigue cracking results are significantly higher than PaveME fatigue cracking predictions. • PaveME fatigue cracking predictions at different as-constructed AV show the expected trend. For four of nine sections, as AV increase, cracking increases. For the other five sections, there was no change in fatigue cracking when AV were varied. Assembling MnROAD and NCAT Validation Datasets Validating predictions of continuous variables examines how the values of the forecasts differ from the values of the observations. The following steps were used to validate the results from all three analysis methods. 1. Validation of continuous predictions included exploratory plots such as scatter plots for Analysis Method 1. This step does not provide quantitative results, and it is prone to individual, subjective interpretation, but it does provide an overall view of the power of estimation of any model. 2. Scatter plot – For Analysis Method 1, this validation method answers the question: How well did the external validation data follow the tendencies of the observed values? 3. Validation of continuous predictions included various model performance parameters. 4. Scatter plot – Plots the model prediction values against the observed values. Answers the question: How well did the estimated model values correspond to the observed values? (Good and Hardin, 2012; Mayer and Buttler, 1993) 5. Box plot – Plots to display the range of data falling between the 25th and 75th percentiles, the horizontal line inside the box showing the median value, and the whiskers showing the complete range of the data. Box plots answer the question: How well did the distribution of forecast model values correspond to the distribution of observed values? (Good and Hardin, 2012; Mayer and Buttler, 1993) 6. Mean error – Answers the question: What is the average forecast model error?

99 7. Each score obtained in step 2 was compared to the same results obtained during the development of the model in Analysis Methods 2 and 3. The apparent performance of the model on this new dataset was expected not to be better than the performance in the model development dataset. Deviation criteria such as 10% difference from the model scores and F-Test of the MSEs were used to establish the level of validation and applicability of the models for estimation purposes. (Montgomery, 2009) MnROAD MnROAD is a pavement research facility made up of various materials and pavements owned and operated by the Minnesota Department of Transportation. Located near Albertville, Minnesota, MnROAD has over 50 unique test sections on the following two roadway segments: a 3.5-mile (I-94) mainline and a 2.5-mile low volume road. MnROAD routinely collects detailed pavement performance data and utilizes pavement sensors located in each test section. Original construction occurred from 1991 until 1994. The test cells were sequentially numbered 1 through 46 with more cells added between 1997 through 2006. New cell additions during Phase I were either designated with a previous number or assigned an available number below 99. MnROAD’s most recent data collection processes and data can be accessed online and is also summarized in Table 2- 34. Table 2-34. MnROAD’s available databases. Layer information Layer widths and thickness for both lanes and shoulders. Lab Testing (HMA, base) Online data from materials sampling, lab procedures, and summary results of quality control testing. Field monitoring (ruts, ride, texture, cracks, strength, etc.) Online data along with procedures used to collect pavement distress, surface characteristics, and other non-destructive testing data. Weather To record the weather conditions at the MnROAD facility, several on- site weather stations have been utilized since 1990. Missing data is often supplemented with a nearby weather station located at the Buffalo, Minnesota regional airport. Traffic Online data for mainline (I-94) and low volume road traffic including collection descriptions. In order to determine the ESAL values for test pavements, traffic volume and axle loading data recorded by WIM sensors are used. Table 2-35 shows the 37 available MnROAD sections used in this validation that were originally constructed as flexible pavements containing several asphalt concrete layers and sections that were milled and/or overlaid using asphalt concrete layers. This table excludes sections with open graded mixtures placed as surface layers. Table 2-35. MnROAD flexible pavement test sections used for validation. Construction Year Sections Roadway 1993 1, 2, 3, 4, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 50, 51, 92, 94, 95, 96, 97 Mainline 1993 25, 26, 27, 29, 30, 31 Low Volume Road 2008 1, 2, 3, 4, 15, 16, 17, 18, 19, 20, 21, 22, 23 Mainline 2008 24, 33, 34, 35 Low Volume Road

100 Daily temperature and precipitation records were available from 1994 to 2016 on MnROAD’s website. Traffic data were found in terms of AADTT and ESALs. Table 2-36 shows the computed annual weather and traffic data using the units required for the study. Construction of MnROAD’s first cycle started in 1993 and finished in 1994. There were no records available with construction dates of each section to account for the increment of time elapsed between construction and open to traffic date. Table 2-36. MnROAD’s weather and traffic data. Year Annual Ave. Daily Temp., °F Annual Precipitation, In Annual Freezing Index Annual Frz / Thaw Days kESALs (Mainline) kESALs (LVR) 1994 42.8 25.42 1155.6 156 264.2* 7.7* 1995 42.2 72.84 1099.4 155 580.0 18.1 1996 39.1 42.47 1567.8 182 533.6 20.1 1997 41.8 20.54 1113.3 177 416.0 16.2 1998 46.6 22.07 696.1 130 541.9 23.2 1999 45.7 12.06 740.0 140 641.0 11.0 2000 44.2 5.08 978.9 131 671.6 20.9 2001 45.9 2.74 852.8 142 714.3 19.1 2002 45.2 3.96 682.8 167 714.2 23.1 2003 44.9 1.62 968.9 141 690.6 16.4 2004 43.8 18.77 817.8 133 435.2 9.2 2005 44.3 25.15 905.0 147 812.0 13.8 2006 46.3 12.71 567.8 135 712.4 15.2 2007 45.4 18.31 981.1 138 760.5 11.6 2008 41.5 17.02 1160.0 153 208.2* 5.6* 2009 43.6 15.35 1044.4 141 773.1 13.8 2010 45.1 13.93 997.2 132 584.2 15.3 2011 44.3 24.94 956.1 149 632.5 12.4 2012 47.2 8.73 642.8 138 719.2 19.2 2013 42.2 15.49 1240.0 166 922.4 13.8 2014 41.7 24.88 1352.2 150 827.4 19.5 2015 46.8 9.44 763.9 128 875.1 18.7 2016 44.5 14.65 1118.7 145 413.7* 15.0* * Partially available due to construction Table 2-37 shows the input data for the MnROAD pavement sections used in this study. Almost all the sections used in this study were new construction and only two sections were in the rehabilitation category. A weighted average of as-constructed AV content and binder content of the entire asphalt layer are also shown in Table 2-37. Details per asphalt layer and other properties used in this study are included in the Appendix.

101 Table 2-37. MnROAD’s pavement sections input data. Cycle Cell Class Type AC Thick. (in) Air Voids (%Gmm) Binder Content (% mix wt) Base Thick. (in) 1993 1 HV-Main New 5.8 7.1 5.3 33 1993 2 HV-Main New 5.5 5.1 5.8 28 1993 3 HV-Main New 5.8 7.3 5.6 33 1993 4 HV-Main New 7.9 7.5 5.3 0 1993 14 HV-Main New 8.9 5.6 5.4 0 1993 15 HV-Main New 8.4 7.1 5.4 0 1993 16 HV-Main New 7.3 8.1 5.1 28 1993 17 HV-Main New 6.9 7.7 5.4 28 1993 18 HV-Main New 7.5 5.7 5.8 21 1993 19 HV-Main New 7.4 6.4 6.1 28 1993 20 HV-Main New 7.1 6.2 6.0 28 1993 21 HV-Main New 6.7 5.3 5.8 23 1993 22 HV-Main New 7.2 6.3 5.3 18 1993 23 HV-Main New 7.3 7.8 5.6 7 1993 25 LV New 5.3 7.5 5.6 0 1993 26 LV New 5.8 8.5 5.8 0 1993 27 LV New 3.1 8.0 6.0 11 1993 29 LV New 3.5 7.3 5.7 6 1993 30 LV New 3.6 8.3 5.3 12 1993 31 LV New 3.8 9.1 5.1 16 2008 24 LV New 3.0 5.3 5.4 4 2008 33 LV New 3.8 6.0 5.3 12 2008 34 LV New 3.6 6.6 5.4 12 2008 35 LV New 3.9 6.5 5.4 12 2008 1 HV-Main Rehab 1.5 3.5 5.3 33 2008 2 HV-Main New 3.0 7.1 5.3 12 2008 3 HV-Main New 3.0 7.8 4.8 8 2008 4 HV-Main New 3.0 6.1 5.1 17 2008 15 HV-Main Rehab 3.0 6.3 5.3 0 2008 16 HV-Main New 5.0 5.6 5.2 24 2008 17 HV-Main New 5.0 6.3 5.2 24 2008 18 HV-Main New 5.0 6.0 5.7 24 2008 19 HV-Main New 5.0 6.0 5.3 24 2008 20 HV-Main New 5.0 5.7 5.5 24 2008 21 HV-Main New 5.0 3.8 5.6 24 2008 22 HV-Main New 5.0 4.9 5.7 24 2008 23 HV-Main New 5.0 6.0 5.7 24 With regards to assembling the performance dataset, three files were used to build the ride dataset and one file was used to build the cracking datasets. The difference in the ride data files reflected the use of different technologies applied by MnROAD throughout the years. One file contained all the cracking information that was collected using visual inspections records (cracking maps).

102 Figures 2-53 and 2-54 show typical performance curves of the mainline ride quality for MnROAD’s 1993 and 2008 cycles, respectively. Steeper curves were observed during the 1993 cycle compared to the 2008 cycle. This behavior seemed to be associated with more rutting and thermal cracking occurring in the 1993 cycle. In addition, extrapolation at 10 years of IRI values were needed in the 2008 cycle since the testing of most sections was finalized in 2015 or 2016. Figure 2-53. Example of MnROAD’s 1993 cycle typical ride performance curve. Figure 2-54. Example of MnROAD’s 2008 cycle typical ride performance curve. Figures 2-55 and 2-56 show typical rutting performance curves of the mainline for MnROAD’s 1993 cycle. It was found that at year eight, crack sealing and slurry sealing was applied on most 1993 sections. As observed in Figure 2-55, these treatments significantly affected rutting performance. Therefore, any rutting data for sections that such treatments were applied were removed from the analysis. As observed in Figure 2-56, it was necessary to extrapolate rut depth at year 10 for most 1993 sections.

103 Figure 2-55. Example of MnROAD’s 1993 cycle typical rutting performance. Figure 2-56. Example of MnROAD’s 1993 Cycle Typical Rutting Performance Curve. Table 2-38 shows a summary of the parameters previously defined for Analysis Method 1 of this study on ride quality and rutting. In several cases, there were not enough data points to extrapolate the results or there was not a significant change in the results to quantify an increase in IRI by 25 points with respect to the initial IRI.

104 Table 2-38. Ride quality and rutting at MnROAD. Cycle Cell IRI at 10 Years, in/mile Years to Incr. Initial IRI by 25 Rutting at 4 Years, in Rutting at 6 Years, in Rutting at 8 Years, in Years at 03.-in Rutting 1993 1 170 5 0.34 0.4 0.45 3 1993 2 185 3 0.4 0.4 0.8 2 1993 3 165 4 0.32 0.37 0.73 3 1993 4 260 3 0.41 0.44 0.8 2.5 1993 14 230 4.5 0.27 0.33 0.65 4.7 1993 15 285 4.5 0.23 0.29 0.59 6 1993 16 250 3.4 0.2 0.21 0.58 7 1993 17 200 1.9 0.24 0.29 0.54 7 1993 18 220 2 0.27 0.3 0.6 6 1993 19 185 2.3 0.31 0.41 0.7 4 1993 20 90 8 0.45 0.52 0.58 1.7 1993 21 80 7 0.38 0.5 0.58 2.5 1993 22 120 2.6 0.3 0.37 0.6 4 1993 23 175 3.6 0.5 0.5 0.58 0.8 1993 25 80 8.1 0.34 0.35 0.59 3.8 1993 26 90 Ins. Data 0.4 0.5 0.54 1.8 1993 27 360 2.5 0.55 0.62 0.62 0.7 1993 29 125 8 0.41 0.47 0.8 2 1993 30 195 6 0.42 0.42 0.7 2 1993 31 330 4 0.44 0.45 0.77 2 2008 24 100 12 0.1 0.14 0.21 9 2008 33 90 Ins. Data 0.2 0.25 0.36 7.2 2008 34 118 6.5 0.2 0.27 0.33 7 2008 35 120 10 0.1 0.13 0.16 12 2008 1 160 1.8 0.17 0.21 0.25 9.5 2008 2 65 11 0.16 0.15 0.2 11 2008 3 90 5 0.12 0.13 0.17 14 2008 4 125 3.5 0.22 0.25 0.35 7 2008 15 145 3 0.17 0.19 0.21 20 2008 16 85 10 0.17 0.2 0.23 11.5 2008 17 100 8 0.23 0.25 0.3 8 2008 18 80 12 0.26 0.29 0.3 8 2008 19 95 10 0.2 0.23 0.28 8.5 2008 20 65 12 0.11 0.12 0.13 30 2008 21 50 Ins. Data 0.12 0.13 0.14 30 2008 22 60 Ins. Data 0.11 0.12 0.14 30 2008 23 88 Ins. Data 0.18 0.23 0.28 8.2 Figures 2-57 and 2-58 show typical wheel path cracking performance and transverse cracking performance curves of the mainline, respectively. Wheel path cracking was originally reported by MnDOT as cracking area and it was not indicated if the cracked area was located on the wheel path. Therefore, it was assumed for this validation task that the entire cracking area was located in the wheel path to comply with the HPMS wheel path width (1.0 meters). As observed in Figure 2-58, it was difficult to fit a trend line

105 on the entire dataset; therefore, the majority of the thermal cracking parameters were obtained directly from the plot. Figure 2-57. Example of MnROAD’s typical wheel path cracking performance curve. Figure 2-58. Example of MnROAD’s typical thermal cracking performance curve. Table 2-39 shows a summary of the parameters previously defined for Analysis Method 1 of this study on cracking. In several cases, there were not enough data points to extrapolate the results or there was not a significant change in the results to identify when the thermal cracking initiated.

106 Table 2-39. Cracking parameters at MnROAD. Cycle Cell % Wheel Path Cracking at 10 Years Thermal Crack Initiation Year Thermal Cracking Init+3 Years 1993 1 0 1.4 190 1993 2 0 1.4 257 1993 3 0 1.4 190 1993 4 0 1.4 297 1993 14 0 1.4 266 1993 15 0 1.4 492 1993 16 0 1.6 527 1993 17 0 1.6 295 1993 18 0 1.6 369 1993 19 0 1.6 506 1993 20 0 1.6 283 1993 21 0 1.6 320 1993 22 0 1.4 222 1993 23 0 1.4 218 1993 25 0 9 110 1993 26 54 13.5 Ins. Data 1993 27 Ins. Data 1.4 447 1993 29 4 9.5 194 1993 30 1.4 1.6 98 1993 31 10 9.3 80 2008 24 Ins. Data 2.5 215 2008 33 Ins. Data 6 Ins. Data 2008 34 Ins. Data Ins. Data Ins. Data 2008 35 Ins. Data Ins. Data Ins. Data 2008 1 30 Ins. Data Ins. Data 2008 2 Ins. Data Ins. Data Ins. Data 2008 3 Ins. Data 6.7 Ins. Data 2008 4 Ins. Data Ins. Data Ins. Data 2008 15 Ins. Data 0.3 180 2008 16 45 5.4 Ins. Data 2008 17 45 Ins. Data Ins. Data 2008 18 Ins. Data Ins. Data Ins. Data 2008 19 45 Ins. Data Ins. Data 2008 20 27 5.1 200 2008 21 45 5.3 210 2008 22 Ins. Data Ins. Data Ins. Data 2008 23 Ins. Data Ins. Data Ins. Data NCAT Test Track NCAT’s Test Track is a 1.7-mile oval where research is conducted on experimental asphalt pavements. The track is comprised of 46 200-ft test sections funded as a cooperative project among highway agencies and industry sponsors. Research experiments are based on single test sections or groups of test sections.

107 Test sections available for NCHRP 20-50(18) were classified as new construction (from structural experiments) and rehabilitation sections (from surface mixture experiments). At the NCAT Test Track, performance is monitored on a continuous basis to evaluate rutting, fatigue cracking, roughness, texture, friction, and noise. Structural pavement sections have varying thicknesses that closely resemble real-world pavements. These sections have embedded strain and pressure sensors that analyze pavement response to loads for validation of mechanistic-empirical pavement design procedures. Surface mix performance sections are built on a robust cross-section that limits distresses to the experimental surface layers. Multi- depth temperature probes were installed in each test section. Paired with an onsite automated weather station, these data are used to characterize the performance environment for each experimental section. Each section on the track is subjected to 10 million ESALs of heavy truck traffic applied over a period of two years. Table 2-40 shows a total of 101 available NCAT sections containing several asphalt concrete layers and sections that were milled and/or overlaid using asphalt concrete layers. This table excludes sections with open graded mixtures placed as surface layers. Table 2-40. NCAT’s available flexible pavement test sections. Construction Year Test Section 2000 E1 to E10, N1 to N13, S1 to S9, W1, W2, W6 to W10 2003 E1 to E3, N1 to N8, W3, W6, W9 2006 E5, E6, E7, N1, N2, N5, N8, N9, N10, W3, W4, W5, S2, S6, S7, S8, S11, S12 2009 N5 to N7, N10, N11, S2, S6, S7, S9 to S12 2012 N3, N4, S3, S5, S6, S12, S13 2015 N1, N2, N5, N7, N8, S4 to S6 NCAT’s most recent data collection processes and data can be accessed online. However, there is no centralized database containing all the data from the previous cycles. For each cycle, there is an excel file containing field performance test results, traffic data, and weather data. To obtain other related information, such as mixture properties and other laboratory performance tests, several NCAT reports were used as a reference (Table 2-41). Table 2-41. NCAT’s available database sources. Construction Year Source 2000 NCAT Report 01-02: As-Built Properties of Experimental Sections on the 2000 NCAT Pavement Test Track NCAT Report 02-12: NCAT Test Track Design, Construction, and Performance 2003 NCAT Report 04-01: Design & Instrumentation of the Structural Pavement Experiment at the NCAT Test Track NCAT Report 06-01: Material Properties of the 2003 NCAT Test Track Structural Study NCAT Report 06-05: Phase II NCAT Test Track Results 2006 NCAT Report 09-01: Design, Construction, and Instrumentation of the 2006 Test Track Structural Study NCAT Report 09-06: Mechanistic Characterization of Resilient Moduli for Unbound Pavement Layer Materials NCAT Report 09-08: Phase III NCAT Test Track Findings

108 2009 NCAT Report 12-10: Phase IV NCAT Pavement Test Track Findings NCAT Report 13-02: Physical and Structural Characterization of Sustainable Asphalt Pavement Sections at the NCAT Test Track 2012 NCAT Report 16-04: Phase V (2012-2014) NCAT Test Track Findings Daily temperature and precipitation records were available from years 2000 to 2017. Table 2-42 shows the computed annual weather data collected at the NCAT Test Track. Construction of NCAT’s first cycle started in 2000 and it has been re-built every three years. Table 2-42. NCAT’s weather data. Year Annual Ave. Daily Temp., °F Annual Precipitation, in Annual Freezing Index, °F-days Annual Frz/Thaw Days 2000 62.2 37.4 55.0 28 2001 63.6 42.4 20.6 19 2002 63.9 70.9 4.4 29 2003 63.0 67.9 11.1 29 2004 63.7 48.7 4.4 25 2005 63.2 65.8 3.9 25 2006 65.4 52.1 18.9 16 2007 65.4 67.1 1.1 24 2008 63.9 40.1 39.4 34 2009 64.0 65.0 7.8 29 2010 64.1 33.8 16.7 55 2011 64.6 40.1 41.1 28 2012 66.4 25.4 0.0 11 2013 63.2 45.5 0.0 24 2014 61.8 40.9 28.9 39 2015 64.8 91.8 33.3 31 2016 65.8 30.8 0.6 21 2017 65.7 51.9 4.4 15 The original Test Track, built in the year 2000, included an improved 12-inch thick subgrade, a 6-inch granular base, a combination of three asphalt treated base layers for 20 inches total, and 4 inches of asphalt surface mixture (Figure 2-59). As of 2018, close to 50% of the sections still have the original structure where only the top layer has been rehabilitated several times. Starting in 2003, several of the tangent sections were redesigned and rebuilt with asphalt layers ranging from 5 to 9 inches thick on top of granular bases ranging from 6 to 9 inches. Figure 2-60 shows an example of pavement cross sections built after 2003. Table 2-43 shows a summary of NCAT’s new and rehabilitated pavement sections per research cycle. Details per asphalt layer and other properties used in this study are also included in the Appendix.

109 Figure 2-59. NCAT Test Track original pavement layers. Source: NCAT Report 9-01 Figure 2-60. Example of new sections built after 2003. 4 inch – Experimental asphalt surface (each section) 6 inch – Asphalt base (PG 76 -22) 9 inch – Asphalt base (PG 67 -22) 5 inch – Permeable asphalt base 6 inch – Aggregate base 12 inch – Aggregate subbase (A-2) Subgrade

110 Table 2-43. Summary of NCAT’s pavement sections. Type Cycle AC-Layer Thickness, in AC-Layer Voids, % Binder Content, % Min Max Min Max Min Max New 2000 3.0 6.6 4.9 7.9 3.8 8.0 2003 4.7 8.8 5.8 7.1 4.5 4.8 2006 7.0 14.1 5.3 7.3 3.2 5.0 2009 5.6 8.9 6.3 7.5 4.8 5.9 2012 3.9 6.4 5.4 6.5 4.9 5.7 2015 5.5 7.1 5.2 7.7 4.6 5.5 Rehab 2003 4.0 4.2 3.0 7.9 5.8 8.2 2006 3.3 4.7 3.1 8.0 4.4 7.5 2009 4.0 4.0 6.9 7.0 5.2 6.0 2012 4.0 4.0 5.8 5.8 6.4 6.4 2015 4.0 4.0 5.0 5.0 6.4 6.4 Traffic is applied for about two years until the 10 million ESAL target is achieved. This accelerated full- scale loading is intended to represent 10 years of traffic on a high volume roadway. In order for this data to be used for validation purposes, a time shift from two to ten years (depending on each cycle) was made on every cycle to approximate normal traffic operations. Figure 2-61 shows an example of the shift made on time on a typical rutting performance curve. In addition, as observed in Figure 2-62, the cumulative application of traffic at the NCAT Test Track was quite uniform, expressed as 1 million ESALs per year (on a 10-year shift scenario). The time shift from two to ten years had recognized limitations related to material performance associated with temperature and aging. For example, rutting occurred in the summer months of the two-year accelerated cycle but were not adjusted to the summer months of the extended ten- year validation period. Also, the measured fatigue cracking extended to ten years did not include the influence of ten years of aging. Real Time Cycle 10 Year Shift Figure 2-61. Example of time shift on NCAT rutting performance curves.

111 Figure 2-62. Cumulative ESALs on NCAT sections over an adjusted 10-year cycle. Figure 2-63 shows typical performance curves at the NCAT Test Track. In most cases, little to no change was observed within one cycle. Table 2-44 shows the range of IRI and rut depth parameters per cycle and by type of experiment (new or rehabilitation study). 1 Cycle Section 2 Cycle section Figure 2-63. Example of NCAT section typical ride quality performance.

112 Table 2-44. Summary of NCAT’s ride quality and rutting performance. Type Cycle IRI at 10 Years, in/mile Years to Incr. Initial IRI by 25 Rutting at 8 Years, in Years at 03.-in Rutting Min Max Min Max Min Max Min Max New 2000 35 140 11 20 0.03 0.31 4.1 13 2003 41 350 2.2 16.5 0.12 0.21 4.5 20 2006 90 190 3.1 12 0.08 0.38 4.7 11 2009 40 118 9 20 0.08 0.3 8 19 2012 70 250 3.3 19 0.17 0.4 3.7 8.2 2015 46 70 11.5 11.5 0.05 0.13 Ins. Data Ins. Data Reh. 2003 56 75 8 15 0.14 0.28 10 18 2006 68 300 0.6 15 0.02 0.65 2.5 11 2009 55 85 22 22 0.14 0.14 Ins. Data Ins. Data 2012 65 65 Ins. Data Ins. Data 0.08 0.08 Ins. Data Ins. Data 2015 41 41 Ins. Data Ins. Data 0.04 0.04 Ins. Data Ins. Data Fatigue cracking data was only available for the 2012 and 2015 cycles. Figure 2-64 shows typical wheel path cracking performance curves in the 2012 and 2015 cycles, respectively. Table 2-45 shows wheel path cracking obtained at 6 and 10 years. In most cases, wheel path cracking initiation was captured after year 6 and cracking growth follows an exponential trend. Cracking was originally reported by NCAT as percent cracking of the entire lane and it was not indicated if the cracked area was located on the wheel path. For this study, it was assumed that the entire cracking area was located in the wheel path and the percent cracking was adjusted to comply with the HPMS wheel path width (1.0 meter). Figure 2-64. Example of NCAT section typical wheel path cracking performance curves.

113 Table 2-45. NCAT’s wheel path cracking performance. Cycle ID Type % WP Cracking Year 6 Year 10 2012 N3 New 0 0 2012 N4 New 0 0 2012 S3 Reh 0 0 2012 S5 New Ins. Data Ins. Data 2012 S6 New 21.9 40 2012 S12 New 0 0 2012 S13 New 34.8 45 2015 N1 New 0 22 2015 N2 New 0 6 2015 N5 New 0 5 2015 N7 New 0 20 2015 N8 New 0.1 17 2015 S4 Reh 0 0 2015 S5 New 0 0 2015 S6 New 0 0 Validation of Analysis Method 1 Using MnROAD and NCAT Datasets MnROAD is located in climatic region 1 and the validation dataset contained sections that associated with two of the traffic-structural number subgroups established for Analysis Method 1. In this case, only new construction sections were included in the analysis since there were only two sections found as rehabilitation sections. Low volume sections met the LOW-THIN subgroup criteria, as shown in Figure 2-65, for rutting at four years. However, MnROAD’s dataset shows an increase in rutting with an increase in as-constructed AV, contrary to the trend followed by LTPP dataset. Some of MnROAD’s sections met the HIGH-THK subgroup criteria. As shown in Figure 2-66, MnROAD’s dataset also followed an opposite trend compared to LTPP dataset. 18 out of 35 sections at MnROAD did not fall into an Analysis Method 1 subgroup. As shown in Figure 2-67, these 18 sections were classified as HIGH-MED and an increase in rutting with an increase in as-constructed AV can be observed. For two of the MnROAD validation subgroups it should be noted that the range of as-constructed AV does not correspond with the LTPP subgroup. For CL1-LOW- THIN there are no MnROAD values below 6%. For CL1-HIGH-THIN there are no LTPP values below 6%.

114 Figure 2-65. MnROAD versus LTPP rutting tendencies CL1-LOW-THIN. Figure 2-66. MnROAD versus LTPP rutting tendencies CL1-HIGH-THK. Figure 2-67. MnROAD rutting tendencies CL1-HIGH-MED.

115 There was no clear wheel path cracking trend observed from LTPP data subgroups. The same case was observed with MnROAD’s subgroup datasets as shown in Figure 2-68 for LOW-THIN sections, Figure 2- 69 for HIGH-THK sections, and Figure 2-70 for the HIGH-MED sections. These results were attributed to the significant scatter in cracking performance. Figure 2-68. MnROAD versus LTPP wheel path cracking tendencies CL1-LOW-THIN. Figure 2-69. MnROAD versus LTPP wheel path cracking tendencies CL1-HIGH-THK.

116 Figure 2-70. MnROAD wheel path cracking tendencies CL1-HIGH-MED. Thermal cracking performance was also characterized by high variability as observed in the scatter plots. For MnROAD sections in the LOW-THIN and HIGH-MED subgroups, a clear trend was not identified (Figure 2-71 and Figure 2-73). MnROAD’s HIGH-THK subgroup data followed an opposite trend compared to the LTPP dataset (Figure 2-72). Figure 2-71. MnROAD versus LTPP thermal cracking tendencies CL1-LOW-THIN.

117 Figure 2-72. MnROAD versus LTPP thermal cracking tendencies CL1-HIGH-THK. Figure 2-73. MnROAD thermal cracking tendencies CL1-HIGH-MED. With regards to ride quality, low volume sections at MnROAD showed an increase in IRI at 10 years and a decrease in the time to increase IRI by 25 in/mi as the as-constructed AV increase (Figure 2-74). For this subgroup, the LTPP dataset did not show any defined trend. The high volume and thick pavement subgroup sections at MnROAD followed the observed trend of the LTPP subgroup data with an increase in IRI as the as-constructed AV increased (Figure 2-75). The remaining sections classified as HIGH-MED subgroup followed the same trend as the other MnROAD subgroups as shown in Figure 2-76.

118 Figure 2-74. MnROAD versus LTPP ride tendencies CL1-LOW-THIN. Figure 2-75. MnROAD versus LTPP ride tendencies CL1-HIGH-THK. Figure 2-76. MnROAD ride tendencies CL1-HIGH-MED.

119 NCAT’s Test Track is located in the climatic region 4 (wet, no freeze) and the NCAT dataset contained sections that met one of the traffic-structural number subgroups established for Analysis Method 1. 21 of the new construction sections at the NCAT Test Track were associated with the HIGH-MED subgroup. The remaining new and rehabilitated sections were classified as HIGH-THIK and HIGH-LOW according to the subgroup categories. Figure 2-77 shows a slight increase in rutting at four years as the as-constructed AV increase on the NCAT dataset while the LTPP dataset did not show a clear trend. The remaining NCAT sections did not follow a clear trend as shown in Figure 2-78 for HIGH-THK new construction sections, in Figure 2-79 for HIGH-LOW new construction sections, and Figure 2-80 for HIGH-THK rehabilitated sections. Figure 2-77. NCAT versus LTPP rutting tendencies CL4-HIGH-MED. Figure 2-78. NCAT rutting tendencies CL4-HIGH-THK.

120 Figure 2-79. NCAT rutting tendencies CL4-HIGH-LOW. Figure 2-80. NCAT rutting tendencies CL4-HIGH-THK (rehabilitation). The amount of wheel path cracking data was very limited at the NCAT Test Track. When comparing HIGH-MED new constructions sections, neither the NCAT nor LTPP dataset showed a clear trend due to variability in the results (Figure 2-81). On the other hand, HIGH-THK new construction sections showed a slight decrease in wheel path cracking at 10 years as the as-constructed AV increased (Figure 2-82).

121 Figure 2-81. NCAT versus LTPP wheel path cracking tendencies CL4-HIGH-MED. Figure 2-82. NCAT wheel path cracking tendencies CL4-HIGH-THK. With regards to ride quality, HIGH-MED new construction sections at the NCAT Test Track did not show a clear trend for the measured IRI at 10 years or in the increment of time to increase IRI by 25 in/mi (Figure 2-83). No trend was observed on the HIGH-THK sections in terms of IRI, but the increment of time to measure 25 in/mi higher than the initial IRI value tended to shorten (Figure 2-84). No clear trend was observed on sections classified as HIGH-LOW as shown in Figure 2-85. HIGH-THK rehabilitated sections only show an increase in the time to increment IRI by 25 in/mi as the as-constructed AV increase (Figure 2-86).

122 Figure 2-83. NCAT versus LTPP ride tendencies CL4-HIGH-MED. Figure 2-84. NCAT ride tendencies CL4-HIGH-THK. Figure 2-85. NCAT ride tendencies CL4-HIGH-LOW.

123 Figure 2-86. NCAT ride tendencies CL4-HIGH-THK (rehabilitation). Validation of Analysis Method 2 Using MnROAD and NCAT Datasets Input data from MnROAD and NCAT were used to predict performance using the Analysis Method 2 models and analyze the trend of observed performance versus predicted performance. Only new construction sections were included in the analysis since there were an insufficient number of MnROAD and NCAT rehabilitation sections. Figure 2-87 shows the complete observed dataset and the predicted values to estimate ride quality at 6 and 10 years. Both increments of performance time were used to assemble a reasonable size dataset for examining validation. As shown, there was poor prediction at values of IRI above 100 in/mi for both LTPP and validation datasets. These results were expected since 90 percent of the LTPP dataset showed IRI values below 100 in/mi; therefore, extrapolation of higher performance results was beyond the model range. The Mean Square Error (MSE) of the validation dataset was 3276, five times higher than the LTPP model MSE of 550. An F-test was used to compare variance of error values from the validation dataset and the LTPP model dataset as a statistical approach to determine the power of estimation of the model over the validation dataset (Table 2-46). The null (H0) F-hypothesis test was defined as the equality of MSEs. Rejection of the null hypothesis means that there is evidence to conclude that MSEs are different at a significance level of 0.05. As expected, the results indicated a significant difference in variance obtained from the model and validation errors (F-value > F-critical). Box plots were used to compare the distribution of model prediction values that correspond to the distribution of observed LTPP values and to compare the distribution of model prediction data with measured external validation data. Figure 2-88 shows that LTPP observed and predicted IRI had similar distributions; however, validation IRI data had more scattered values and its corresponding predicted distribution was significantly different. This confirmed the need for selecting the validation data that matched the distribution of IRI values found in the model’s data. Validation data was filtered to avoid extrapolation and reduce uncertainty of the predicted results. Filtering of the data included removing sections that did not meet input criteria established for Analysis Method 1 subgroups and were out of the range of input variables present in the LTPP database. For instance, almost 50% of the NCAT sections (climate zone 4) were chararcterized by having high SN (above 6.4) and heavy traffic (over 10 million ESALs). There were an insufficient number of LTPP dataset sections in the Analysis Method 1 subgroup to examine the trend.

124 When the validation dataset was filtered to avoid extrapolation, a better comparison of the ride quality was observed and the validation data was more in line with the model data (Figure 2-89). The computed MSE of the filtered validation dataset was significantly reduced to 996, which was 81% higher than the model’s MSE. In this case, rejection H0 was obtained at a significance level of 0.05. However, this rejection was marginal and for practical purposes the model seems to work on the filtered validation data. Figure 2-87. Observed versus predicted validation ride data (new pavements). Figure 2-88. Box-plots of observed and predicted ride data (new pavements).

125 Figure 2-89. Observed versus predicted validation ride data, filtered (new pavements). Table 2-46. New pavements ride F-test results. Parameter IRI Observed LTPP IRI Validation Data IRI Validation Filtered MSE 549.9 3276.1 995.8 Observations 1765 223 73 DF 1764 222 72 F-value 5.957 1.811 F-Critical 1.209 1.362 Result Reject H0 Reject H0 Figure 2-90 shows the relationship between observed and predicted IRI data for rehabilitated pavements. Validation data seems to follow the general trend obtained for LTPP data. However, the results of an F- Test in Table 2-47 indicated that MSEs were significantly different. These results were expected since almost all the validation NCAT data came from sections with SN above 6.4 and with heavy traffic, which were conditions not heavily represented in the LTPP dataset. No further filtering and analysis was performed since only two validation sections were considered usable to avoid extrapolation. A comparison of the error distributions of the model predictions using LTPP data and validation data is shown in Figure 2-91. The majority of error values computed on LTPP data were concentrated between ±10, while the error distribution of the validation data varied from -30 to 40. This difference in error ditributions helped explain the results of the F-Test.

126 Figure 2-90. Observed versus predicted validation ride data (rehabilitated pavements). Table 2-47. Rehabilitated pavements ride F-test results. Parameter IRI Observed LTPP IRI Validation Data MSE 468.0 959.5 Observations 1598 47 DF 1597 46 F-value 2.050 F-Critical 1.458 Result Reject H0 Figure 2-91. Ride data error distribution. Figure 2-92 shows the relationship between observed and predicted rutting for LTPP data and a similar relationship for the validation data after filtering. In this case, the validation MSE was 0.044, three times higher than the model’s MSE (0.011). It can be observed that predicted results were highly scattered and the validation data is not entirely contained within the model’s data. An F-Test indicated that the validation and model MSE’s are statistically different (Table 2-48).

127 Figure 2-92. Observed versus predicted validation rutting, filtered (new pavements). Table 2-48. New pavements rutting F-test results. Parameter Rutting Observed LTPP Rutting Validation Data MSE 0.011 0.044 Observations 1688 98 DF 1687 97 F-value 4.199 F-Critical 1.312 Result Reject H0 Figure 2-93 shows the relationship between observed and predicted rutting (rehabilitated sections) for LTPP data and the validation data after filtering. In this case, the validation prediction MSE was 0.014 and it was 27% higher than the LTPP data model MSE (0.011). The LTPP predicted results were highly scattered and the validation data is not entirely contained within the model’s data. An F-Test indicated that the validation and model variances of the errors can be equal (Table 2-49). Figure 2-93. Observed versus predicted validation rutting, filtered (rehabilitated pavements).

128 Table 2-49. Rehabilitated pavements rutting F-test results. Parameter Rutting Observed LTPP Rutting Validation Data MSE 0.011 0.014 Observations 1249 63 DF 1248 62 F-value 1.355 F-Critical 1.395 Result Confirm H0 Figure 2-94 shows the relationship between observed and predicted wheel path cracking for the LTPP data and the validation data after filtering. In this case, the MSE of the validation prediction was 92, 29% lower than the LTPP model MSE (130.5). The predicted results were highly scattered and the validation data is clearly contained within the model’s data. An F-Test indicated that the validation and model variances of the errors can be equal (Table 2-50). There were an insufficient number of validation sections to validate the wheel path cracking on rehabilitated sections. Figure 2-94. Observed versus predicted validation wheel path cracking, filtered (new pavements). Table 2-50. New pavements wheel path cracking F-test results. Parameter Wheel Path Cracking Observed LTPP Wheel Path Cracking Validation Data MSE 130.5 92.4 Observations 1533 32 DF 1532 31 F-value 1.412 F-Critical 1.774 Result Confirm H0 Figure 2-95 shows the relationship between observed and predicted transverse cracking for LTPP data and the validation data (MnROAD only) after filtering. In this case, the validation MSE was 16% higher than the model’s MSE. The LTPP predicted results were slighly scattered and the validation data were mostly contained within the model’s data. An F-Test indicated that the validation and model variances of the errors can be equal (Table 2-51).

129 Figure 2-95. Observed versus predicted validation transverse cracking, filtered (new pavements). Table 2-51. New pavements transverse cracking F-test results. Parameter Transverse Cracking Observed LTPP Transverse Cracking Validation Data MSE 334256 388464 Observations 199 22 DF 198 21 F-value 1.162 F-Critical 1.760 Result Confirm H0 Validation of Analysis Method 3 Using MnROAD and NCAT Data Once again, input data from MnROAD and NCAT were used to estimate and analyze the trend of observed performance versus predicted performance. Figure 2-96 shows a correlation plot between observed and predicted ride quality data. The ride data comes from the observed IRI values at 10 years of trafficking at MnROAD and NCAT. These data were also filtered to show data points within the parameters established for the LTPP dataset. Figure 2-97 shows observed and predicted rutting at 6 years. These results were also filtered. Figure 2-98 shows observed and predicted wheel path cracking at 10 years and Figure 2-99 shows observed and predicted thermal cracking at 10 years at MnROAD. In all cases, it was observed that the external validation data points were more scattered compared to the LTPP-ANN models data. These results suggest that the ANN approach was a good fit for the LTPP dataset but that the prediction of new performance values is questionable.

130 Figure 2-96. Observed versus ANN predicted validation ride data. Figure 2-97. Observed versus ANN predicted validation rutting data. Figure 2-98. Observed versus ANN predicted validation wheel path cracking data.

131 Figure 2-99. Observed versus ANN predicted validation transverse cracking data. F-tests were also used to compare variance of error values from the validation dataset and the model dataset as a statistical approach to determine the power of estimation of the model over the validation dataset (Table 2-52). Mean square errors of the external validation datasets were significantly higher than the neural network models on all cases. Therefore, the results of F-tests indicated that the validation and model variances of the errors are different. Table 2-52. Artificial neural network F-test results MSE DF F-value F-Critical Result IRI Model (training data) 11.6 829 Validation-predicted 11691 105 1007.9 1.31 Reject H0 Validation-filtered predicted 8272 26 713.1 1.63 Reject H0 Rutting Model (training data) 0.001 884 Validation-predicted 0.25 100 241.4 1.32 Reject H0 Validation-filtered predicted 0.29 26 283.5 1.63 Reject H0 WP Cracking Model (training data) 26.0 846 Validation-predicted 1351 30 51.9 1.58 Reject H0 Validation-filtered predicted 791 3 30.4 3.13 Reject H0 T. Cracking Model (training data) 1711 774 Validation-predicted 156261 17 91.3 1.79 Reject H0 Validation-filtered predicted 65922 4 38.5 2.80 Reject H0

Next: Chapter 3: Findings and Applications »
Investigating the Relationship of As-Constructed Asphalt Pavement Air Voids to Pavement Performance Get This Book
×
MyNAP members save 10% online.
Login or Register to save!
Download Free PDF

Several controlled laboratory studies have shown that air voids (AV) can have a large effect on the performance of asphalt pavements. AVs that are either too high or too low can cause a reduction in pavement life.

The TRB National Cooperative Highway Research Program's NCHRP Web-Only Document 299: Investigating the Relationship of As-Constructed Asphalt Pavement Air Voids to Pavement Performance determines the effect of in-place AVs on the performance of asphalt concrete (AC) pavements.

The document also has supplemental appendices that are available by request to Ed Harrigan. They include data sets for LTPP, Pavement ME Design Validation, MnROAD Validation, and NCAT Validation.

  1. ×

    Welcome to OpenBook!

    You're looking at OpenBook, NAP.edu's online reading room since 1999. Based on feedback from you, our users, we've made some improvements that make it easier than ever to read thousands of publications on our website.

    Do you want to take a quick tour of the OpenBook's features?

    No Thanks Take a Tour »
  2. ×

    Show this book's table of contents, where you can jump to any chapter by name.

    « Back Next »
  3. ×

    ...or use these buttons to go back to the previous chapter or skip to the next one.

    « Back Next »
  4. ×

    Jump up to the previous page or down to the next one. Also, you can type in a page number and press Enter to go directly to that page in the book.

    « Back Next »
  5. ×

    To search the entire text of this book, type in your search term here and press Enter.

    « Back Next »
  6. ×

    Share a link to this book page on your preferred social network or via email.

    « Back Next »
  7. ×

    View our suggested citation for this chapter.

    « Back Next »
  8. ×

    Ready to take your reading offline? Click here to buy this book in print or download it as a free PDF, if available.

    « Back Next »
Stay Connected!