U.S. Department of Transportation
Federal Highway Administration
1200 New Jersey Avenue, SE
Washington, DC 20590
202-366-4000


Skip to content
Facebook iconYouTube iconTwitter iconFlickr iconLinkedInInstagram

Federal Highway Administration Research and Technology
Coordinating, Developing, and Delivering Highway Transportation Innovations

 
REPORT
This report is an archived publication and may contain dated technical, contact, and link information
Back to Publication List        
Publication Number:  FHWA-HRT-11-045    Date:  November 2012
Publication Number: FHWA-HRT-11-045
Date: November 2012

 

Performance Testing for Superpave and Structural Validation

CHAPTER 4. MECHANISTIC-EMPIRICAL ANALYSIS OF ALF TEST LANES

INTRODUCTION

Mechanistic-empirical analyses of the ALF test lanes were completed for several reasons. The first was to ascertain whether or not construction variability of the HMA and unbound base and subgrade layers influenced the rankings of the measured performance, with the exclusion of lane 6 rutting, as described in chapter 3. The second purpose was to assess the absolute predictive and relative ranking capabilities of mechanistic-empirical pavement performance prediction models within the NCHRP 1-37A MEPDG methodology.(1)

The following three different types of MEPDG analyses were conducted:

Fully parametric sensitivity studies that could evaluate the influence of individual aspects of construction on performance (e.g., the influence of only HMA layer thickness compared to the ideal 4- or 5.8-inch (100- or 150-mm) design thickness) were not conducted.

It must be recognized that TPF-5(019) includes polymer-modified asphalts that are central to the design of the experiments. However, not by design, the LTPP database used to provide the national calibration of NCHRP 1-37A and subsequent versions of the MEPDG did not include a large number of polymer modified asphalt data points. Accurate performance predictions are not necessarily guaranteed or expected.

FWD ANALYSIS OF UNBOUND LAYER MODULI

Central to any mechanistic-empirical pavement analysis are linear elastic material properties assumed to represent the pavement layers. A series of FWD tests were conducted after the CAB reconstruction was complete and again after the AC layers were placed. The locations of FWD testing were at the midpoint of each pair of survey plates at the centerline of the ALF wheel path. Since 8 survey plates were installed in each of the 4 test sites, a total of 7 locations were measured by FWD for each site and 28 for each lane. At each location, load levels of 6,000, 9,000, 12,000, and 16,000 lbf (27, 40, 52, and 70 kN) were targeted. Three replicates were recorded at each load level. This resulted in a total of 12 tests at each location, 84 tests for each test site, and 4,032 tests for all pavement sections. All FWD tests were performed by a Dynatest® 2000 unit with a neoprene-padded load plate 11.8 inches (0.3 m) in diameter. The sensors were spaced at distances of 0, 8, 12, 18, 24, 36, 48, 60, and -12 inches (0, 0.203, 0.305, 0.457, 0.610, 0.914, 1.219, 1.524, and -0.305 m) from the center of load plate.

Composite Modulus on CAB

The composite modulus calculated using only the center deflection sensor and stress offers a means to quickly quantify the variability of the granular base construction without elaborate back-calculation routines. This does not represent realistic engineering properties of the different layers. Figure 58 shows the variation in composite modulus, with stiffer responses toward lanes 1 and 2 in the lower station numbers (sites 1 and 2) and a mild variation in modulus elsewhere. Two locations with extreme differences were chosen to explore the variation with a back-calculation scenario that assumed bedrock and saturated subgrade. The back of lane 2 (sites 1 and 2) had a stiff response while lane 11 had a softer response. The back-calculated modulus of CAB was 16,099 and 11,168 psi (111 and 77 MPa) from lanes 2 and 11, respectively. The back-calculated modulus of the subgrade was 9,572 and 7,251 psi (66 and 50 MPa) from lanes 2 and 11, respectively.

This line chart shows lanes 1 through 12 on the x-axis and composite modulus on the y-axis from the falling weight deflectometer (FWD) on top of a crushed aggregate base (CAB). Two series of line charts are represented, and each site has four points represented from sites 1 and 2 in the first series and sites 3 and 4 in the second series. Error bars are plotted with the data points and represent one standard deviation.
1 kPa = 0.145 psi

Figure 58. Graph. Variation in composite modulus from FWD on top of CAB.

These specific moduli and overall variation in composite modulus were held for consideration until a more thorough back-calculation of the pavement layers was completed after construction of the AC layers. The CAB was exposed to loss or gain of moisture that could contribute to variations in the stiffness. Variations of stiffness could also be due to magnification of true variations in stiffness without the presence of a generally stiffer AC layer above. Furthermore, root mean square error (RMSE) of the deflection basins fit during these back-calculations was rather large, between 8 and 22 percent.

Back-Calculated Modulus of Pavement Structure

Another series of FWD tests was conducted after the AC layers were placed. Pavement temperature at the mid-depth of the asphalt layer was measured during the FWD tests using a thermometer inserted in a hole drilled in each asphalt layer and filled with oil. The pavement temperature was recorded at the beginning and end of each site. The structural configuration in figure 16 provided a starting point for trial configurations that were used to determine a suitable scheme for the back-calculation. The depth to bedrock of the locality was reported between 25 and 30 ft (7.5 to 9 m) based on observations during construction of geotechnical test pits nearby. Saturated subgrade layers can influence back-calculation computations by mimicking a stiff layer. This was considered in trial configurations as follows:

The depth to bedrock used in the analysis was 25 ft (7.62 m). When a saturated subgrade layer was placed above the bedrock, 30 percent of the total subgrade layer, or 6.99 ft (2.13 m), was assumed saturated while the other 70 percent, or 15.8 ft (4.83 m), was not a fixed subgrade modulus. Bedrock modulus was assumed fixed at 507,632 psi (3,500 MPa), and the saturated subgrade modulus was fixed at 50,763 psi (350 MPa). The full set of stations in lanes 2 and 11 were analyzed using the EVERCALC back-calculation software.(55) Results from the back-calculation are shown in figure 59 and figure 60. The analysis indicates that both four-layer systems produced nearly identical modulus. The five-layer system calculated moduli for the base, which is stiffer than both four-layer systems, and also calculates softer subgrade layers. All layer configuration cases calculated modulus that varies along the length of the stationing and is similar in magnitude. Sometimes the base is stiffer than the subgrade, and sometimes the subgrade is stiffer than the CAB.

This graph shows a plot of falling weight deflectometer (FWD) back-calculated crushed aggregate base (CAB) modulus from various trial layer configurations on the y-axis and the longitudinal station number on the x-axis. Example data from lanes 2 and 11 are given with a total of six series. Three series are plotted for each lane, and each series represents three different back-calculation scenarios.
1 MPa = 145 psi

Figure 59. Graph. FWD back-calculated CAB modulus from various trial layer configurations.

 

The graph shows a plot of falling weight deflectometer (FWD) back-calculated subgrade modulus from various trial layer configurations on the y-axis and the longitudinal
1 MPa = 145 psi

Figure 60. Graph. FWD back-calculated subgrade modulus from various trial layer configurations.

These exploratory analyses of the two lanes helped determine the layer configuration and back-calculation scheme for the remaining sites. The bedrock depth and modulus as well as saturated subgrade depth and modulus were implemented in all subsequent back-calculations. FWD data were taken at all 4 sites of all 12 lanes so that variations in stiffness could be accounted for in the mechanistic-empirical pavement analyses, but as a matter of practicality, only the center station of each site was back-calculated for subsequent analyses. Two back-calculation programs were used to analyze the same set of data for a more diverse analysis, since optimization algorithms can be different from software to software. The two software packages used were EVERCALC and MODCOMP/MODTAG. Seed modulus for the asphalt layers was determined by interpolating measured indirect tension (IDT) resilient modulus measured on lab-compacted, plant-produced mixtures at 50, 68, 86, and 104 °F (10, 20, 30, and 40 °C). This is shown schematically in figure 61.

This graph shows eight curves for each asphalt mixture that plot the variation of indirect tension (IDT) resilient modulus on the y-axis and temperature on the x-axis. Modulus decreases exponentially with an increase in temperature. A vertical and horizontal arrow illustrate interpolation used to obtain a falling weight deflectometer back-calculation seed modulus at any given temperature.
1 kPa = 0.145 psi
°F = 1.8(°C) + 32

Figure 61. Graph. Variation of IDT resilient modulus with temperature.

Back-calculated moduli of the asphalt layers are summarized in table 41, back-calculated moduli of the base layers are summarized in table 42, and back-calculated moduli of the subgrade layers are summarized in table 43. In each table, the values from EVERCALC and MODCOMP/MODTAG are provided along with the percent difference relative to the EVERCALC values. The back-calculated HMA modulus was at most 69 percent larger or 32 percent smaller than the seed value, but there was no bias to the back-calculated HMA modulus relative to the seed modulus, which fluctuated between larger and smaller.

Table 41. Modulus back-calculation results for the HMA layers.

Lane

Site

HMA Temperature (°C)

EVERCALC (MPa)

MODTAG/ MODCOMP (MPa)

Percent Difference

2

1

11.1

14,000

15,625

-10

2

11.1

14,000

16,050

-13

3

11.7

12,470

14,000

-11

4

11.7

9,139

9,948

-8

3

1

11.7

7,169

6,805

5

2

11.7

9,637

8,805

9

3

11.7

6,662

5,855

14

4

13.3

7,931

8,093

-2

4

1

12.8

9,491

8,088

17

2

12.8

6,668

6,280

6

3

11.3

8,193

7,125

15

4

12.8

7,604

7,975

-5

5

1

26.5

3,132

3,318

-6

2

27.2

2,781

3,033

-8

3

29.3

2,456

2,818

-13

4

29.3

3,199

3,630

-12

6

1

8.7

12,914

12,600

2

2

9.3

9,537

9,740

-2

3

9.8

8,782

10,165

-14

4

11.7

7,802

9,600

-19

8

1

10.0

11,850

14,950

-21

2

13.9

9,247

11,350

-19

3

10.4

7,808

9,803

-20

4

14.1

7,100

9,220

-23

9

1

28.1

1,235

1,733

-29

2

26.1

1,091

1,418

-23

3

26.1

999

1,420

-30

4

26.7

942

1,290

-27

10

1

13.3

8,789

9,453

-7

2

13.0

9,950

12,175

-18

3

13.0

5,965

6,968

-14

4

13.9

5,821

6,825

-15

11

1

13.9

5,455

5,735

-5

2

13.9

6,748

6,875

-2

3

13.3

4,380

6,068

-28

4

13.3

5,074

6,965

-27

12

1

13.0

4,617

5,633

-18

2

12.2

4,976

5,200

-4

3

12.0

5,173

6,913

-25

4

12.0

4,452

4,715

-6

°F = 1.8(°C) + 32
1 MPa = 145 psi

Table 42. Modulus back-calculation results for the CAB.

Lane

Site

EVERCALC (MPa)

MODTAG/ MODCOMP (MPa)

Percent Difference

2

1

92

90

2

2

74

66

11

3

68

57

19

4

70

64

10

3

1

55

60

-8

2

54

62

-12

3

51

59

-14

4

61

59

4

4

1

59

71

-16

2

56

59

-4

3

61

70

-14

4

54

51

7

5

1

56

53

5

2

62

57

9

3

54

48

11

4

60

54

11

6

1

60

64

-5

2

64

63

2

3

51

41

23

4

80

63

27

8

1

114

55

105

2

114

66

73

3

143

91

58

4

110

56

95

9

1

66

45

45

2

60

47

27

3

63

45

41

4

74

55

36

10

1

79

66

20

2

88

46

93

3

98

71

38

4

107

76

41

11

1

100

90

11

2

99

94

5

3

134

78

70

4

125

70

79

12

1

96

66

45

2

118

107

10

3

129

78

65

4

134

120

12

Minimum

51

41

-16

Maximum

143

120

105

Average

82

66

26

Std. deviation

28

17

COV (percent)

34

26

1 MPa = 145 psi
— Indicates that data were not provided because they were not relevant.

COV = Coefficient of variation.

Table 43. Modulus back-calculation results for the subgrade.

Lane

Site

EVERCALC (MPa)

MODTAG/ MODCOMP (MPa)

Percent Difference

2

1

80

75

7

2

69

68

2

3

79

81

-1

4

72

71

1

3

1

71

65

8

2

68

62

9

3

62

57

9

4

65

62

4

4

1

72

65

11

2

66

63

5

3

67

61

10

4

64

62

2

5

1

62

61

2

2

69

68

1

3

57

57

0

4

61

61

0

6

1

73

68

7

2

79

75

5

3

60

63

-4

4

72

76

-6

8

1

96

113

-15

2

81

90

-10

3

91

99

-8

4

84

99

-15

9

1

65

71

-8

2

66

69

-5

3

72

80

-10

4

76

83

-8

10

1

79

78

1

2

81

93

-13

3

91

95

-4

4

97

102

-5

11

1

85

83

3

2

84

80

5

3

99

111

-11

4

97

111

-13

12

1

83

88

-6

2

80

79

1

3

96

107

-10

4

92

92

1

Minimum

57

57

-15

Maximum

99

113

11

Average

77

79

-1

Std. Deviation

12

16

COV (percent)

15

21

1 MPa = 145 psi
— Indicates that data were not provided because they were not relevant.

COV = Coefficient of variation.

Base and subgrade modulus have notable variation across all 48 sites in the 12 ALF lanes. The coefficients of variation (COVs) of the base modulus from the two back-calculation software programs were 34 and 26 percent, and the COVs of the subgrade were 15 and 21 percent from EVERCALC and MODCOMP/MODTAG, respectively. The average base modulus was 11,893 psi (82 MPa) from EVERCALC and 9,572 psi (66 MPa) from MODCOMP/MODTAG. The average subgrade modulus was 11,168 psi (77 MPa) from EVERCALC and 11,458 psi (79 MPa) from MODCOMP/MODTAG. Overall, EVERCALC calculated moduli of the base and subgrade that tended to be closer to one another and MODCOMP/MODTAG calculated base moduli that tended to be less stiff than the subgrade. The maximum and minimum RMSE reported from the EVERCALC back-calculation were 3.5 and 0.8 percent, with an average and standard deviation of 1.5 and 0.65 percent. The average RMSE from MODCOMP/MODTAG was larger, about 4.01 percent.

Evaluation of Effective Loading Frequency from FWD

The back-calculated HMA moduli in table 41 at the particular temperatures of the FWD measurement were overlaid onto the fit dynamic modulus master curves |E*| (see next section). The mixes from the thicker 5.8-inch (150-mm) HMA lanes were used because the construction allowed tall enough cores to be taken for laboratory measurements of axial modulus. Time-temperature superposition was used to calculate a reduced frequency based on the measured temperature and assumed input effective frequency of the FWD measurement. Then, the input frequency was adjusted and optimized to find the best agreement between the fit master curve of the field mix cores and the FWD modulus back-calculation. The overall effective frequency of the FWD pulses was about 8.2 Hz. Figure 62 through figure 66 illustrate how the measured FWD moduli compare to the direct measurement of the cores’ dynamic modulus for each mix. There is notable scatter because both the back-calculated moduli from EVERCALC and MODCOMP/MODTAG were used, but the relationship appears reasonable when the values are viewed in the inlaid log-log scale figure. When each individual mix/lane was optimized, the effective frequencies ranged between about 4 and 16 Hz.

This graph shows a plot of lane 8 performance grade (PG) 70-22 control asphalt mixture dynamic modulus on the y-axis in arithmetic scale and loading frequency on the x-axis in log scale. The plot is focused on a relatively small range of loading frequency and modulus, showing seven individual data points from falling weight deflectometer (FWD) back-calculated asphalt modulus. The fit dynamic modulus master curve measured on field cores is plotted with the data points for comparison between laboratory and field modulus. An inset in the lower right corner shows the entire laboratory-measured master curve and FWD data points over the full range of frequency and modulus.
1 MPa = 145 psi

Figure 62. Graph. Lane 8 PG70-22 asphalt mixture dynamic modulus.

 

This graph shows a plot of lane 9 styrene-butadiene-styrene (SBS) 64-40 asphalt mixture dynamic modulus on the y axis in arithmetic scale and loading frequency on the x-axis in log scale. The plot is focused on a relatively small range of loading frequency and modulus, showing seven individual data points from falling weight deflectometer (FWD) back-calculated asphalt modulus. The fit dynamic modulus master curve measured on field cores is plotted with the data points for comparison between laboratory and field modulus. An inset in the lower right corner shows the entire laboratory-measured master curve and FWD data points over the full range of frequency
1 MPa = 145 psi

Figure 63. Graph. Lane 9 SBS 64-40 asphalt mixture dynamic modulus.

 

This graph shows a plot of lane 10 air-blown asphalt mixture dynamic modulus on the y-axis in arithmetic scale and loading frequency on the x-axis in log scale. The plot is focused on a relatively small range of loading frequency and modulus, showing seven individual data points from falling weight deflectometer (FWD) back-calculated asphalt modulus. The fit dynamic modulus master curve measured on field cores is plotted with the data points for comparison between laboratory and field modulus. An inset in the lower right corner shows the entire laboratory-measured master curve and FWD data points over the full range of frequency and modulus.
1 MPa = 145 psi

Figure 64. Graph. Lane 10 air-blown asphalt mixture dynamic modulus.

 

The graph shows a plot of lane 11 linear grafted styrene-butadiene-styrene (SBS-LG) asphalt mixture dynamic modulus on the y-axis in arithmetic scale and loading frequency on the x-axis in log scale. The plot is focused on a relatively small range of loading frequency and modulus, showing seven individual data points from falling weight deflectometer (FWD) back-calculated asphalt modulus. The fit dynamic modulus master curve measured on field cores is plotted with the data points for comparison between laboratory and field modulus. An inset in the lower right corner shows the entire laboratory-measured master curve and FWD data points over the full range of frequency and modulus.
1 MPa = 145 psi

Figure 65. Graph. Lane 11 SBS-LG asphalt mixture dynamic modulus.

 

The graph shows a plot of lane 12 terpolymer asphalt mixture dynamic modulus on the y-axis in arithmetic scale and loading frequency on the x-axis in log scale. The plot is focused on a relatively small range of loading frequency and modulus, showing seven individual data points from falling weight deflectometer (FWD) back-calculated asphalt modulus. The fit dynamic modulus master curve measured on field cores is plotted with the data points for comparison between laboratory and field modulus. An inset in the lower right corner shows the entire laboratory-measured master curve and FWD data points over the full range of frequency and modulus.
1 MPa = 145 psi

Figure 66. Graph. Lane 12 terpolymer asphalt mixture dynamic modulus.

Assessment of Unbound Layer FWD Back-Calculation with Multiple Depth Deflectometers

The reasonableness of the back-calculated moduli of the unbound base and subgrade was checked against vertical deflections measured from within the pavement structure (as opposed to only the surface deflections from FWD) using multiple depth deflectometers (MDDs). Two MDDs were installed in lane 4 (4-inch (100-mm) SBS-LG) and lane 11 (5.8-inch (150-mm) SBS-LG), for a total of four MMDs. Each MDD has linear variable differential transformers (LVDTs) to measure the deformation between the MDD head box at the surface and the LVDT anchors at different pavement depths. The LVDT anchors were installed at the top of the CAB layer, in the middle of the CAB, and at the top of the subgrade. The bottom of the MDDs were anchored 10 ft (3 m) below the pavement surface. A schematic of an MDD is shown in figure 67.

This schematic illustration shows a vertical pavement cross section with depth, including anchor points and measurement locations of a multidepth deflectometer (MDD) instrumentation installed within a borehole.
Courtesy of CTL Group

Figure 67. Illustration. Vertical pavement cross section showing anchor points and measurement locations of MDD.

MDD responses were measured by both FWD loading and ALF rolling wheel loading. FWD response to impact loading was measured before ALF loads were applied at three locations for each set of MDDs: directly above the MDD and 12 and 24 inches (300 and 600 mm) offset from the MDD. During testing, the pavement temperature was measured at the surface by a handheld non-contact infrared device and at the middle of the HMA layer by a thermometer in a drilled hole filled with oil. The temperature of the FWD measurements was 66 °F (19 °C). The FWD load magnitudes were 9,000, 12,000, and 16,000 lbf (40, 53, and 70 kN). In contrast to the FWD loading, the ALF wheel loading was under fixed load and varying temperature. The ALF wheel load was 10,000 lbf (44 kN), and the temperatures were 50, 66, 82, and 147 °F (10, 19, 28, and 64 °C). The locations of the ALF wheel load relative to the MDD were directly above and offset at 8 and 12 inches (200 and 300 mm).

A schematic of the forward-calculation scheme based on back-calculation showing the various depths, thicknesses, and stiffness is provided in figure 68. Two different moduli for each layer were used in the layered elastic predictions representing practical and typical variation in modulus. The HMA moduli were taken from the laboratory-measured dynamic modulus |E*| master curves from cores or estimates of cores at frequencies of 5 and 15 Hz to cover the range of effective FWD frequencies identified in the previous section (8 Hz). The two moduli chosen for the unbound layer properties were taken from the extreme softest and stiffest back-calculated moduli of each lane in table 42 and table 43, whether from EVERCALC or MODCOMP/MODTAG.

This illustration shows a side-by-side schematic layout of lanes 4 and 11 pavement layer configurations used for the forward-calculation scheme of multidepth deflectometer (MDD) instrumentation response. The following five pavement layers are shown (with thickness and modulus values labeled): hot mix asphalt, crushed aggregate base, weather affected subgrade, saturated subgrade, and bedrock hard bottom.
°F = 1.8(°C) + 32
1 MPa = 145 psi
1 mm = 0.039 inches

Figure 68. Illustration. Layout of lanes 4 and 11 pavement layer configuration for forward-calculation scheme of MDD instrumentation response.

The measured and predicted MDD deformations from FWD loading are shown in table 44 and table 45 as well as in figure 69 and figure 70. Overall, the two MDD measurements have some variability, but the general trends of smaller deformations deeper in the pavement, farther away from the load, and with smaller loads is captured. The error bars on the predicted MDD deformations represent the standard deviation about the average of the eight different combinations of moduli for the pavement layers used in the predictions. The predicted deformations appear to be less sensitive to modulus than the measured variation in MDD response. The variability in the measured MDD deformations does not appear to be unreasonable given the variability in the back-calculated moduli within and between lanes, as shown in table 42 and table 43.


Table 44. MDD peak deflections in mm for lane 4 during FWD loading at 66 °F (19 °C).

Load (kN)

Offset (mm)

Top of CAB

Mid of CAB

Top of Subgrade

MDD-1

MDD-2

Layered Elastic Predicted

MDD-1

MDD-2

Layered Elastic Predicted

MDD-1

MDD-2

Layered Elastic Predicted

Average

Std. Dev.

Average

Std. Dev.

Average

Std. Dev.

40

0

0.96

0.85

0.79

0.066

0.60

0.44

0.47

0.034

0.42

0.18

0.31

0.022

300

0.55

0.40

0.53

0.042

0.42

0.33

0.39

0.027

0.31

0.23

0.28

0.020

600

0.25

0.23

0.28

0.022

0.22

0.22

0.25

0.018

0.19

0.19

0.21

0.016

53

0

1.31

1.16

1.05

0.088

0.81

0.68

0.63

0.045

0.60

0.43

0.41

0.030

300

0.72

0.44

0.70

0.056

0.59

0.36

0.51

0.036

0.45

0.24

0.37

0.027

600

0.34

0.33

0.37

0.030

0.31

0.32

0.33

0.024

0.27

0.27

0.28

0.021

70

0

1.77

1.33

1.39

0.116

1.16

0.84

0.83

0.059

0.83

0.55

0.55

0.039

300

0.96

0.55

0.93

0.074

0.81

0.45

0.68

0.048

0.62

0.29

0.49

0.035

600

0.46

0.48

0.49

0.039

0.44

0.46

0.44

0.032

0.38

0.40

0.36

0.027

1 kN = 225 lbf
1 mm = 0.039 inches

Table 45. MDD peak deflections in mm for lane 11 during FWD loading at 66 °F (19 °C).

Load (kN)

Offset (mm)

Top of CAB

Mid of CAB

Top of Subgrade

MDD-1

MDD-2

Layered Elastic Predicted

MDD-1

MDD-2

Layered Elastic Predicted

MDD-1

MDD-2

Layered Elastic Predicted

Average

Std. Dev.

Average

Std. Dev.

Average

Std. Dev.

40

0

0.47

0.38

0.42

0.024

0.33

0.34

0.29

0.012

0.29

0.40

0.22

0.008

300

0.38

0.27

0.32

0.013

0.35

0.26

0.25

0.009

0.28

0.25

0.20

0.007

600

0.18

0.16

0.20

0.006

0.18

0.15

0.18

0.005

0.17

0.16

0.16

0.005

53

0

0.64

0.53

0.56

0.031

0.46

0.48

0.39

0.016

0.38

0.56

0.29

0.010

300

0.48

0.38

0.42

0.017

0.44

0.38

0.34

0.012

0.35

0.36

0.26

0.009

600

0.26

0.22

0.27

0.008

0.25

0.20

0.24

0.007

0.25

0.21

0.21

0.006

70

0

0.87

0.70

0.75

0.042

0.68

0.63

0.52

0.021

0.55

0.74

0.39

0.014

300

0.67

0.52

0.57

0.023

0.60

0.52

0.45

0.016

0.48

0.50

0.35

0.012

600

0.36

0.29

0.36

0.010

0.35

0.28

0.32

0.009

0.34

0.29

0.28

0.008

1 kN = 225 lbf
1 mm = 0.039 inches

This graph shows nine three-by-three subplots of measured and predicted multidepth deflectometer (MDD) peak deflection data given in table 44 for lane 4 during the falling weight deflectometer (FWD) loading at 66 °F (19 °C). Each subplot shows the vertical MDD deflection on the y-axis and horizontal distance from the applied FWD load on the x-axis. Subplots arranged in columns show the variation in deflection due to FWD drop load, and subplots arranged in rows show the variation of depth within the pavement. Each subplot includes two replicate measurements as a dashed line and measured deflection as a solid line.
1 mm = 0.039 inches

Figure 69. Graph. Measured and predicted MDD peak deflection data for lane 4 during FWD loading at 66 °F (19 °C).

 

This graph shows nine three-by-three subplots of measured and predicted multidepth deflectometer (MDD) peak deflection data given in table 45 for lane 11 during the falling weight deflectometer (FWD) loading at 66 °F (19 °C). Each subplot graphs the vertical MDD deflection on the y-axis and horizontal distance from the applied FWD load on the x-axis. Subplots arranged in columns show the variation in deflection due to FWD drop load, and subplots arranged in rows show the variation of depth within the pavement. Each subplot includes two replicate measurements as a dashed line and measured deflection as a solid line.
1 mm = 0.039 inches

Figure 70. Graph. Measured and predicted MDD peak deflection data for lane 11 during FWD loading at 66 °F (19 °C).

In terms of predicted accuracy, the deformations in lane 4 are better overall than in lane 11. The predictions at all depths, offsets, and load levels are within an agreeable range of the measured deformations in lane 11. The far-field response at 23 inches (600 mm) lateral offset is predicted quite well in lanes 4 and 11 at all depths, which suggests that the deeper back-calculated moduli, fixed modulus of the hard bottom, and weather-affected subgrade are reasonable. The deeper deformations toward the center of the load in lane 11 are not predicted as accurately and suggest the pavement structure layer is softer than what was simulated from the back-calculations.

The measured and predicted MDD deformations from ALF wheel loading are shown in table 46 and table 47 as well as in figure 70 through figure 73. Overall, the trends are quite similar to the FWD loading with respect to lane 4 being better predicted than lane 11 and the far-field responses being predicted well. There is less accuracy in the predicted values, and the measured MDD deflections appear to be a bit more erratic than with FWD loading. Based on this analysis, the back-calculated moduli across the ALF sites are reasonable given that the focus of the experiment was in the HMA layers (not the unbound layers) and the vertical deformation at the bottom of the CAB was always captured well.

Table 46. MDD peak deflections in mm for lane 4 during ALF rolling wheel loading at 10,000 lbf (44 kN).

Temp. (°C)

Offset (mm)

Top of CAB

Middle of CAB

Top of Subgrade

MDD-1

MDD-2

Layered Elastic Predicted

MDD-1

MDD-2

Layered Elastic Predicted

MDD-1

MDD-2

Layered Elastic Predicted

Average

Std. Dev.

Average

Std. Dev.

Average

Std. Dev.

10

0

0.82

0.57

0.76

0.05

0.57

0.37

0.48

0.03

0.44

0.24

0.32

0.02

200

0.59

0.55

0.63

0.05

0.47

0.37

0.44

0.03

0.29

0.28

0.31

0.02

300

0.45

0.49

0.54

0.04

0.41

0.36

0.40

0.03

0.26

0.26

0.29

0.02

19

0

0.87

0.63

0.89

0.07

0.64

0.42

0.53

0.04

0.39

0.27

0.35

0.02

200

0.68

0.56

0.71

0.05

0.45

0.40

0.48

0.03

0.37

0.27

0.33

0.02

300

0.51

0.52

0.59

0.04

0.39

0.38

0.43

0.03

0.31

0.26

0.31

0.02

28

0

0.98

0.90

1.06

0.10

0.73

0.61

0.58

0.04

0.41

0.41

0.37

0.03

200

0.70

0.79

0.80

0.07

0.50

0.56

0.52

0.04

0.40

0.44

0.34

0.03

300

0.53

0.67

0.62

0.05

0.44

0.50

0.46

0.03

0.34

0.39

0.32

0.02

64

0

1.23

1.21

1.92

0.21

0.83

0.82

0.79

0.06

0.52

0.56

0.44

0.03

200

0.86

0.98

1.02

0.10

0.66

0.68

0.65

0.05

0.46

0.50

0.41

0.03

300

0.68

0.73

0.64

0.06

0.55

0.50

0.53

0.04

0.41

0.35

0.37

0.03

°F = 1.8(°C) + 32
1 mm = 0.039 inches

Table 47. MDD peak deflections in mm for lane 11 during ALF rolling wheel loading at 10,000 lbf (44 kN).

Temp. (°C)

Offset (mm)

Top of CAB

Middle of CAB

Top of Subgrade

MDD-1

MDD-2

Layered Elastic Predicted

MDD-1

MDD-2

Layered Elastic Predicted

MDD-1

MDD-2

Layered Elastic Predicted

Average

Std. Dev.

Average

Std. Dev.

Average

Std. Dev.

10

0

0.94

0.51

0.58

0.02

0.65

0.51

0.37

0.01

0.42

0.53

0.26

0.01

200

0.64

0.45

0.48

0.02

0.49

0.45

0.34

0.01

0.39

0.45

0.25

0.01

300

0.36

0.36

0.40

0.01

0.34

0.36

0.31

0.01

0.32

0.35

0.23

0.01

19

0

1.03

0.65

0.67

0.04

0.65

0.62

0.40

0.01

0.42

0.68

0.27

0.01

200

0.64

0.57

0.53

0.02

0.50

0.57

0.36

0.01

0.38

0.58

0.26

0.01

300

0.38

0.43

0.43

0.01

0.37

0.43

0.32

0.01

0.35

0.43

0.24

0.01

28

0

1.24

0.83

0.80

0.05

0.86

0.77

0.44

0.02

0.54

0.89

0.29

0.01

200

0.93

0.69

0.59

0.03

0.75

0.70

0.39

0.01

0.54

0.73

0.27

0.01

300

0.49

0.57

0.45

0.02

0.51

0.58

0.34

0.01

0.40

0.58

0.25

0.01

64

0

2.31

1.84

1.34

0.06

1.68

1.21

0.57

0.02

1.21

1.96

0.35

0.01

200

1.41

1.31

0.71

0.03

1.30

1.03

0.48

0.01

1.25

1.19

0.32

0.01

300

0.53

0.76

0.46

0.01

0.81

0.77

0.39

0.01

0.81

0.63

0.29

0.01

°F = 1.8(°C) + 32
1 mm = 0.039 inches

This graph shows six three-by-two subplots of measured and predicted multidepth deflectometer (MDD) peak deflection data given in table 46 for lane 4 accelerated loading facility (ALF) rolling wheel peak deflections
1 mm = 0.039 inches

Figure 71. Graph. Measured and predicted MDD peak deflection data for lane 4 ALF rolling wheel peak deflections at 50 and 66 °F (10 and 19 °C).

 

This graph shows six three-by-two subplots of measured and predicted multidepth deflectometer (MDD) peak deflection data given in table 46 for lane 4 accelerated loading facility (ALF) rolling wheel peak deflections (in millimeters) at 82 and 147 °F (28 and 64 °C). Each subplot graphs the vertical MDD deflection on the y-axis and horizontal distance from the applied ALF rolling wheel load on the x-axis. Subplots arranged in columns show the variation in deflection due to temperature, and subplots arranged in rows show the variation of depth within the pavement. Each subplot includes two replicate measurements as a dashed line and measured deflection as a solid line.
1 mm = 0.039 inches

Figure 72. Graph. Measured and predicted MDD peak deflection data for lane 4 ALF rolling wheel peak deflections at 82 and 147 °F (28 and 64 °C).

 

This graph shows six three-by-two subplots of measured and predicted multidepth deflectometer (MDD) peak deflection data given in table 47 for lane 11 accelerated load facility (ALF) rolling wheel peak deflections (in millimeters) at 50 and 66 °F (10 and 19 °C). Each subplot graphs the vertical MDD deflection
1 mm = 0.039 inches

Figure 73. Graph. Measured and predicted MDD peak deflection data for lane 11 ALF rolling wheel peak deflections at 50 and 66 °F (10 and 19 °C).

 

This graph shows six three-by-two subplots of measured and predicted multidepth deflectometer (MDD) peak deflection data given in table 47 for lane 11 accelerated load facility (ALF) rolling wheel peak deflections (in millimeters) at 82 and 147 °F (28 and 64 °C). Each subplot graphs the vertical MDD deflection on the y-axis and horizontal distance from the applied ALF rolling wheel load on the x-axis. Subplots arranged in columns show the variation in deflection due to temperature, and subplots arranged in rows show the variation of depth within the pavement. Each subplot includes two replicate measurements as a dashed line and measured deflection as a solid line.
1 mm = 0.039 inches

Figure 74. Graph. Measured and predicted MDD peak deflection data for lane 11 ALF rolling wheel peak deflections at 82 and 147 °F (28 and 64 °C).

Seasonal Monitoring of Pavement Sections with FWD

Lanes 4 and 11 were monitored over 26 months as part of a focused seasonal monitoring program using only FWD loading. At least one test per month was conducted, and, at times, a separate test was carried out before or after heavy rainstorms or periods of wet weather. The data were analyzed to determine the sensor offset location that indicated there was little to no change in deflection regardless of the season or weather characteristics. The variation in deflection throughout the year was evident at the sensor at 24 inches (61 cm), while the sensor at 36 inches (91 cm) was essentially unaffected. This indicates that the depth to a saturated subgrade that can be said to be unaffected by weather was shallower than assumed.

DIRECT MEASUREMENT OF ASPHALT LAYER MODULUS

IDT Resilient Modulus of Asphalt

As described in the previous section, resilient modulus of the HMA mixtures was measured in IDT. Plant-produced mixture sampled during construction was compacted to 7 percent air voids and tested at 50, 68, 86, and 104 °F (10, 20, 30, and 40 °C). Results are shown in table 48 and figure 61.

Table 48. IDT resilient modulus of plant produced material.

Lane/Mixture

IDT Resilient Modulus (psi)

10 °C

20 °C

30 °C

40 °C

Lane 1, CR-AZ

1,167,298

498,752

305,418

112,158

Lane 5, CR-TB

1,292,216

562,813

245,616

215,253

Lane 7, fiber

1,928,210

944,047

506,297

195,421

Lanes 8 and 2, PG70-22

1,656,403

811,647

355,733

241,517

Lane 9, SBS 64-40

516,065

197,292

93,501

59,914

Lanes 10 and 3, air blown

1,517,831

902,838

269,850

157,291

Lanes 11 and 4, SBS-LG

1,087,988

460,965

231,473

78,880

Lanes 12 and 6, terpolymer

1,021,153

418,016

237,114

90,983

1 psi = 6.89 kPa
°F = 1.8(°C) + 32

Dynamic Modulus of Field Cores and Plant- and Laboratory-Produced Mixtures

The companion database for the research project contains the individual replicate data for dynamic modulus and phase angles for the field cores and the plant- and laboratory- produced mixtures.

Field Cores

Field cores with a diameter of 4 inches (100 mm) were taken from the thicker 5.8-inch (150-mm) sections in lanes 8 through 12 in 2003, about 1 year after construction. Field cores had a rough bottom from the interface with the CAB. This rough portion was trimmed smooth to create samples that were shorter than the 5.8-inch (150-mm) height specified by the AMPT protocol.(56) The air void contents of these cores are given in table 49. Only the three tallest samples of the cores were selected for |E*|. The heights of those samples were typically between 4.3 and 4.5 inches (110 and 115 mm). Nonetheless, these samples provided an opportunity to measure the as-constructed modulus of the test sections. The samples were fitted with gauge points glued over the center 2.9-inch (75-mm) portion of the sample and were characterized for |E*| using spacers in the AMPT to accommodate the slightly shorter samples.

Table 49. Air void content of field cores for dynamic modulus.

Lane

Binder

Average Air Void Content (percent)

8

PG70-22

4.8

9

SBS 64-40

5.2

10

Air blown

3.9

11

SBS-LG

5.2

12

Terpolymer

4.5

The temperatures at which the specimens were characterized deviated slightly from the standard temperatures of 40, 70, 100, and 130 °F (4.4, 21.1, 37.8, and 54.4 °C) in the protocol. Instead, the ALF field cores were characterized at 66, 88, 115, and 136 °F (19, 31, 46, and 58 °C) at frequencies of 0.1, 0.5, 1, 5, 10, and 20 Hz. The low-temperature features of the AMPT were malfunctioning at the time the data were collected. The dynamic modulus |E*| master curves of the field cores are shown in figure 75 in log-log scale to highlight the variation in the low modulus region and in figure 76 in semilog scale to highlight the variation in the stiffer moduli. Only the range of measured values is shown; the master curves are not extrapolated in any way. Several observations can be made regarding the stiffness variation with temperature and frequency of the mixtures in their field density conditions.

This graph shows |E*| dynamic modulus on the y-axis in log scale and reduced frequency on the x-axis in log scale. Five master curves for each 5.8-inch (150-mm)-thick lane illustrate an S shaped master curve with increasing stiffness with increasing reduced frequency.
1 MPa = 145 psi

Figure 75. Graph. |E*| dynamic modulus for field cores versus reduced frequency in log-log scale.

 

 This graph shows |E*| dynamic modulus on the y-axis in arithmetic scale and reduced frequency on the x-axis in log scale. Five master curves for each 5.8-inch (150-mm)-thick lane illustrate an S-shaped master curve with increasing stiffness with increasing reduced frequency.
1 MPa = 145 psi

Figure 76. Graph. |E*| dynamic modulus for field cores versus reduced frequency in semilog scale.

In the extreme high-temperature low-frequency region where rutting was primarily induced, the lane 12 terpolymer and lane 9 SBS 64-40 mixtures had the lowest and similar stiffness. Lane 9 (SBS 64-40) had the overall lowest stiffness of all mixtures. The lane 8 PG70-22 and lane 11 SBS-LG mixtures had the next highest stiffness in the high-temperature low-frequency region of the master curve. The stiffest mixture in the extreme high-temperature low-frequency region of the master curve is the lane 10 air-blown mixture.

In the intermediate 66 °F (19 °C) temperature region, which is shown in the right portion of the curves in figure 75 and figure 76, the stiffest mixture was found in lane 8 (PG70-22) followed closely by lane 10 (air blown). The softest mixture was found in lane 9 (SBS 64-40). Lane 12 (terpolymer) and lane 11 (SBS-LG) were similar and had a stiffness between the extremes.

Plant-Produced Mixtures

Loose mixture was sampled in pails during construction and compacted to a target density of 7 percent. The average air void content of the specimens tested is shown in table 50. The lane 7 fiber and lane 1 CR-AZ mixtures were not tested. The plant-produced mixtures were tested at the same temperatures and frequencies at which the field cores were tested: 66, 88, 115, and 136 °F (19, 31, 46, and 58 °C) at frequencies of 0.1, 0.5, 1, 5, 10, and 20 Hz. The low-temperature features of the AMPT were malfunctioning at the time the data were collected. The dynamic modulus |E*| master curves of the plant-produced mixtures are shown in figure 77 in log scale to highlight the variation in the low modulus region and in figure 78 in semilog scale to highlight the variation in the stiffer moduli. Only the range of measured values is shown; the master curves are not extrapolated in any way.

Table 50. Air void content of plant-produced mixture for dynamic modulus.

Lane

Binder

Average Air Void Content (percent)

2 and 8

PG70-22

6.8

3 and 10

Air blown

6.2

4 and 11

SBS-LG

6.8

5

CR-TB

7.0

6 and 12

Terpolymer

6.0

9

SBS 64-40

6.5

This graph shows |E*| dynamic modulus on the y-axis in log scale and reduced frequency on the x-axis in log scale. Six master curves for each plant-produced mixture illustrate an S-shaped master curve with increasing stiffness with increasing
1 MPa = 145 psi

Figure 77. Graph. |E*| dynamic modulus for plant-produced mixtures versus reduced frequency in log-log scale.

 

This graph shows |E*| dynamic modulus on the y-axis in arithmetic scale and reduced frequency on the x-axis in log scale. Six master curves for each plant-produced mixture illustrate an S-shaped master curve with increasing stiffness with increasing reduced frequency.
1 MPa = 145 psi

Figure 78. Graph. |E*| dynamic modulus for plant-produced mixtures versus reduced frequency in semilog scale.

There appears to be less variation in stiffness of the plant-produced mixtures than was observed in the field cores in the extreme high-temperature low-frequency region. In this region, lane 5 CR-TB is the stiffest mixture followed by lanes 3 and 10 (air blown) and lanes 2 and 8 (PG70-22), which appear to be nearly identical to lanes 6 and 12 (terpolymer), and then lanes 4 and 11 (SBS-LG). Similar to the field cores, the lane 9 SBS 64-40 is the softest mixture.

Despite differences in air void content, the order of the mixtures’ moduli at the intermediate temperature range on the left side of the figure is not very different from that of the field cores. The stiffer mixtures are lanes 2 and 8 (PG70-22), lanes 3 and 10 (air blown), and lane 5 (CR-TB). Intermediate stiffness mixtures are lanes 6 and 12 (terpolymer) and lanes 4 and 11 (SBS-LG). The softest mixture is again lane 9 (SBS 64-40).

Lab-Produced Mixtures

Like the plant-produced mixture, the lab-produced mixtures were fabricated to a target density of 7 percent air voids. The average air void content of the lab-produced mixtures is shown in table 51. The lane 7 fiber and lane 1 CR-AZ mixtures were also tested. The low-temperature features of the AMPT were operating properly, and data were collected at 39 °F (4 °C) for all mixtures.

Table 51. Air void content of lab-produced mixture for dynamic modulus

Lane

Binder

Average Air Void Content (percent)

1 top

CR-AZ

2 and 8

PG70-22

7.0

3 and 10

Air blown

6.8

4 and 11

SBS-LG

7.1

5

CR-TB

6.7

6 and 12

Terpolymer

6.7

7

Fiber

9

SBS 64-40

7.4

— Indicates the mixture was not lab produced for this experiment.

Figure 79 and figure 80 show the lab-produced dynamic modulus master curves in log-log and semilog scale, respectively. The two stiffest mixtures in the low-frequency high-temperature range of the master curve at the left side are air blown and CR-TB. The two softest mixtures were SBS 64-40 and CR-AZ. Intermediate stiffness mixtures in the high-temperature low-frequency region were terpolymer, PG70-22, SBS-LG, and fiber. The logarithmic scale emphasizes differences between the mixtures, but their moduli were not extremely different in this range.

This graph shows |E*| dynamic modulus on the y-axis in log scale and reduced frequency on the x-axis in log scale. Eight master curves for each lab-produced mixture illustrate an S-shaped master curve with increasing stiffness with increasing
1 MPa = 145 psi

Figure 79. Graph. |E*| dynamic modulus for lab-produced mixtures versus reduced frequency in log-log scale.

 

This graph shows |E*| dynamic modulus on the y-axis in arithmetic scale and reduced frequency on the x-axis in log scale. Eight master curves for each lab-produced mixture illustrate an S-shaped master curve with increasing stiffness with increasing reduced frequency.
>1 MPa = 145 psi

Figure 80. Graph. |E*| dynamic modulus for lab-produced mixtures versus reduced frequency in semilog scale.

In the intermediate- and low-temperature high-frequency range, the mixtures are more different from one another, as seen more clearly in semilog scale. There are distinct clusters of mixtures where the PG70-22 binder, fiber, and air blown are stiffest. The softest mixture is SBS 64-40. The remaining mixtures, CR-AZ, CR-TB, SBS-LG, and terpolymer, all have similar stiffness.

The phase angle master curves from the dynamic modulus tests for the lab-produced mixtures are shown in figure 81. At the left side of the curve in the high-temperature low-frequency region, the mixtures with the more elastic or lowest phase angles are SBS 64-40, SBS-LG, and terpolymer while the mixtures with the more viscous or highest phase angles are PG70-22, fiber, CR-TB, CR-AZ, and air blown.

This graph shows curves fit to phase angle measured during dynamic modulus test on the y-axis and reduced frequency in the x-axis. Eight series are plotted for each lab-produced mixture, and the overall trend shows a concave down curve with a peak. Individual mixtures show peak magnitudes at different frequency values.

Figure 81. Graph. Curves fit to phase angle measured during dynamic modulus test versus reduced frequency.

The plant-produced laboratory-compacted dynamic modulus and laboratory-batched dynamic modulus are comparable for the PG70-22, air-blown, and SBS-LG mixtures, whereas the plant-produced SBS 64-40, terpolymer, and CR-TB mixtures are stiffer than lab-batched. Cores from the 5.8-inch (150-mm) lane 8 PG70-22 and lane 9 SBS 64-40 sections were stiffer than lab-batched materials but were also more dense. The 5.8-inch (150-mm) lane 10 air-blown cores were less stiff than the lab-produced samples, which were more dense than the cores. Cores from the 5.8‑inch (150-mm) lane 11 SBS-LG and lane 12 terpolymer sections exhibited comparable stiffness to the lab-batched materials, but the cores were also more dense.

MEPDG AND STANDALONE ANALYSES OF ALF PAVEMENT

Several unintended consequences of various features within the MEPDG software would not allow the most faithful simulation of the ALF loading. A surrogate for the MEPDG was utilized that enabled features implemented within the MEPDG to be adjusted or turned off completely. Although it would have been desirable to use the MEPDG specifically, a standalone version of the MEPDG developed by Thyagarajan et al. was implemented within Microsoft Excel®.(57) The MEPDG standalone software was developed to conduct independent analyses for comparison with MEPDG predictions for the purpose of analyzing the consequences of assuming constant tire pressure and constant tire contact area in the MEPDG for strain computations. The software was validated to ensure the predictions of the MEPDG without any chosen adjustments could be reproduced. An example of valid reproduction of the MEPDG is shown in figure 82.(57)

This graph shows rut depth on the y-axis and pavement age on the x-axis. Two series are plotted; one is the rutting predicted by the Mechanistic-Empirical Pavement Design Guide (MEPDG), and the other is from the stand-alone application with load-strain linear proportionality assumption. The two predictions are in good agreement.
>1 inch = 25.4 mm

Figure 82. Graph. Rut depth versus pavement age from MEPDG and standalone application.

ALF Wheel and Tire

The MEPDG offers a special axle configuration option for users to customize and define the number of contact areas, the size of contact area, the wheel load, and the inflation pressure. Pressure is uniform and circular in shape. The ALF’s 425 super-single tire was simulated using the custom axle configuration for rutting with a wheel load of 10,000 lbf (44 kN) and tire inflation pressure of 100 psi (689 kPa). Fatigue loading utilized 16,000 lbf (71 kN) wheel load and 120 psi (827 kPa) inflation pressure.

Wheel wander can also be prescribed with the custom axle configuration by specifying the standard deviation of an ideal normal Gaussian distribution of wheel wander. Zero wheel wander was input for rutting, and the same standard deviation of 5.25 inches (133 mm) programmed into the ALF mechanical loading was used for fatigue.

Another caveat of the special axle configuration wheel loading is that the number of passes per month of the special axle must be input. However, many times, the ALF loading was completed within a month. Naturally, this creates scenarios where there would be very few output data points when ALF loading was simulated faithfully. Instead, approximately 58 data points, or 58 months, were assumed to provide sufficient resolution for all predicted distresses. Therefore, fatigue was programmed to have 5,208 passes per month over 58 months for a total of 302,064 passes. Rutting was programmed to have 700 passes per month over 58 months for a total of 40,600 passes.

A final caveat associated with the custom axle configuration option is that a representative frequency must be specified for all depths. The speed of the tire induces a stress pulse that is distributed within the pavement structure. The net effect is that effective load time gets longer with depth; that is, frequency of the load become smaller with depth. This feature within the normal load spectra portion of the MEPDG appears not to have been implemented in the custom axle configuration option. A frequency of 10 Hz was arbitrarily chosen as representative for all depths for the ALF wheel speed of 11 mi/h (18 km/h). This was considered conservative given an analysis of the ALF tire and speed using the empirical Odemark technique within the MEPDG that determined effective load pulse time and frequency at the top surface to be 18 Hz and the frequency at the bottom of the HMA layers to be 9.1 and 7.3 Hz for 4- and 5.8-inch (100- and 150-mm) sections, respectively, using HMA moduli typical for rutting.(1) When wheel load and moduli typical for fatigue were used in the Odemark technique, the frequency at the top surface was 15 Hz, but the frequency at the bottom of the HMA layers was found to be 4.2 and 3.1 Hz for 4- and 5.8-inch (100- and 150-mm) sections, respectively.

ALF Temperature and Aging

The MEPDG was developed to analyze pavement designs over a realistic lifetime and therefore considers seasonal and daily variations in temperature and moisture. The Enhanced Integrated Climatic Model (EICM) supplies the MEPDG with an input file that contains a complex statistical record of temperature variations throughout the depth of the pavement at various times of day, month, and year. This feature is then used to adjust the moduli of various pavement layers to simulate seasonal effects on pavement mechanics.

The controlled temperatures of the ALF challenged the EICM feature of the MEPDG. Controlled temperatures were simulated by formatting an EICM text input file that forced temperature at all depths and all times of the year to be constant. However, the consequence of the forced temperature for rutting within the MEPDG was that higher temperatures applied over the programmed 58 months artificially stimulated the Global Aging System Model in the MEPDG to increase the stiffness of the HMA layers over time to emulate oxidative aging and hardening. Figure 83 shows an example of this increase in stiffness of the HMA layer from one of the early runs of the MEPDG.

This graph shows asphalt modulus on the y-axis and pavement age on the x-axis. The modulus is output from the Mechanistic-Empirical Pavement Design Guide (MEPDG) and shows an unrealistic and undesired increase in stiffness with age from the aging model.
1 psi = 6.89 kPa

Figure 83. Graph. Asphalt modulus versus pavement age from early run of MEPDG.

Circumventing and eliminating this excessive stiffening behavior within the framework of the MEPDG was attempted by inputting binder viscosity temperature susceptibility parameters, which govern age hardening and have a very low slope. This was unsuccessful and produced erroneous results. The standalone MEPDG surrogate in Microsoft Excel® allowed the stiffening of the modulus with time caused by the Global Aging System Model to be turned off.

Other Caveats of the MEPDG and Standalone

To validate the standalone application for ALF conditions, the distresses predicted over the design period in site 1 (147 °F (64 °C) rutting at 10,000 lbf (44 kN)) of lane 2 were compared with the values predicted by the MEPDG. The aged modulus computed by the MEPDG over the period was used in the standalone application. The bottom-up fatigue cracking and HMA rutting predicted by both the MEPDG and the standalone application compared well. For efficiency in all other sites and lanes, the distress predicted by the standalone application at the first month of the design period was compared with the MEPDG predictions. In the first month, both the MEPDG and standalone analyses used the same HMA modulus.

The distresses (both HMA rutting and bottom-up fatigue cracking) predicted by the standalone application in the first month in site 1 (147 °F (64 °C) rutting at 10,000 lbf (44 kN)) and site 2 (165 °F (74 °C) rutting at 10,000 lbf (44 kN)) of all the lanes compared well with the MEPDG predictions. In site 3 (66 °F (19 °C) fatigue at 16,000 lbf (71 kN)) of all the lanes, the HMA rutting predicted by the standalone application in the first month was higher than the MEPDG predictions. However, the bottom-up fatigue cracking predicted by both analyses matched well. The possible reason behind this may be the heavy load used in site 3. For the given tire inflation pressure (120 psi (827 kPa)), the heavy load corresponds to wider tire contact area. The particular implementation of the layered elastic analysis program JULEA used in MEPDG analyses has some difficulties computing the strain levels in the top of the HMA layer under these conditions. This is not to say JULEA is incorrect, rather, what was computed from the output was likely incorrect. The MEPDG extrapolates the strain computed at the surface and at the depth of 0.148 inches (3.76 mm) multiplied by the contact radius to compute strain levels within this region. The strain computed in this region corresponds to a tensile strain instead of compression that causes rutting. The HMA rutting model has a depth correction factor kz which is mostly negative for the top HMA layer (0.5 inches (13 mm) thick). The standalone application computes the HMA rutting when the product of the correction factor and the strain response is positive. This prevents numerical error while solving the HMA rutting model. The heavy load in site 3 increased the compressive strain at the top sublayer, and, along with the negative correction factor, this resulted in high rutting at the top sublayer. This might be the reason for the high rutting predicted by the standalone application when compared to the MEPDG predictions. The possible solution to this was to change the rutting computation procedure in the standalone application, which may result in no rutting in the top sublayer. Of course, this is only an issue when rutting in the 66 °F (19 °C) fatigue sections is compared and evaluated.

HMA Dynamic Modulus Input to MEPDG

Reflecting As-Built HMA Layer Conditions with Dynamic Modulus |E*| Input

Construction data in chapter 2 showed all 12 ALF lanes were not constructed with identical density. The thinner 4-inch (100-mm) sections in lanes 1–7 were compacted to a slightly less dense state than the mixtures in the thicker 5.8-inch (150-mm) sections in lanes 8–12. Also, there was variation in density within the 4- and 5.8-inch (100- and 150-mm) sections.

Ideally, the in situ material for all 12 lanes would be directly characterized for dynamic modulus |E*| and input into the MEPDG. This was possible for lanes 8–12, as discussed in the previous section. Advantages to this approach are that the effect of field compaction is directly taken into consideration and that it accounts for HMA plant production rather than ideal laboratory batching and compaction (see lime nugget distribution in chapter 2). A number of researchers have found gyratory-compacted HMA mixtures exhibit stiffer and stronger material properties than field compaction. (See references 58–62.) However, the nominal 4-inch (100-mm) thickness of lanes 1–7 prevented cores taken from the ALF pavements to be reasonably characterized in standard |E*| protocols.

The as-built HMA dynamic modulus |E*| input for the 4-inch (100-mm) lanes was estimated as follows. For lane 2 (PG70-22), lane 3 (air blown), lane 4 (SBS-LG), and lane 6 (terpolymer), the difference in air void content between 5.8-inch (150-mm) field cores and the density of the mixture of interest in the 4-inch (100-mm) test section was determined. Then, a density correction factor was used to adjust the known modulus at a given air void content to a softer or stiffer modulus at a slightly different air void content. For lane 5 (CR-TB), the difference in air void content between the lab-compacted plant-produced materials and the density of the test lane was determined. Then, the correction factor was applied to adjust the stiffness of the lab sample to the field conditions. Modulus alone cannot explain the notably better cracking performance of lane 1 (CR-AZ) and lane 7 (fiber) compared to the other mixtures in the 4-inch (100-mm) sections. In addition, the less modest dynamic modulus |E*| of the gap-graded CR-AZ mixture is more likely a reflection of the tests being ran without confinement as well as different volumetric properties compared to the other dense graded mixtures. The repeated load permanent deformation flow number test did use confinement.

A density correction factor for HMA dynamic modulus used in the described procedure was generated using the Witczak and Hirsch predictive equations for dynamic modulus.(1,63) The variations in air void content, VMA, and VFA properties of the HMA mixtures from construction data were collected as inputs to the two predictive equations. A reference condition of 5 percent air voids was chosen. The predicted moduli from the equations for a wide range of binder viscosity input for the Witczak model and binder |G*| input for the Hirsch model were used to generate the ratio of modulus across all temperatures. It was observed that for every 1 percent increase or decrease in air void content there was a corresponding 5.69 percent decrease or increase in stiffness. The slope of the fit line in figure 84 is 0.0569.

This graph shows the ratio between predicted dynamic modulus at various air void contents relative to a reference condition on the y-axis. The x-axis shows the air void content used in dynamic modulus predictions. Two groups of data points are included, representing two dynamic modulus prediction models. The trend shows a decrease in modulus ratio as air void content increases. A linear regression line is fit through the data points that yields a correction factor developed for density effects on dynamic modulus |E*|.

Figure 84. Graph. Ratio between predicted dynamic modulus at various air void contents relative to a reference condition.

MEPDG |E*| Input Formatting

The MEPDG can import measured |E*| dynamic modulus at temperatures and frequencies determined by the user. The MEPDG then conducts the time and temperature shifting and assembles dynamic modulus master curves. However, the MEPDG requires dynamic modulus at temperatures equal or colder than 14 °F (-10 °C). Also, the MEPDG cannot accept moduli softer than 10,000 psi (69 MPa) (very high temperatures and low frequencies). The raw data and master curves shown in figure 75 through figure 80 did not have data at temperatures as cold as required by the MEPDG and contained isolated moduli that were softer than 10,000 psi (69 MPa). The low-temperature data were synthesized by fitting a Hirsch predictive model for dynamic modulus using the volumetric properties of the mixtures and measured binder |G*| dynamic shear moduli. Very low-temperature dynamic moduli were calculated assuming limiting binder moduli of 290,000 psi (2 GPa). Predicted dynamic moduli at 10 °F (-12 °C) were input along with the measured |E*| data into the MEPDG. An example of the extrapolation and limiting low modulus is shown in figure 85.

This graph shows percent cracking on the y-axis and number of accelerated load facility passes on the x-axis and uses fatigue cracking predicted by the Mechanistic-Empirical Pavement Design Guide (MEPDG) standalone program. The group of curves from the 5.8-inch (150-mm) lanes is smaller than the group of curves from the 4-inch (100-mm) lanes, with limited interspersion between.
1 MPa = 145 psi

Figure 85. Graph. Extrapolated dynamic modulus in log scale versus reduced frequency.

The combined measured and low-temperature extrapolated |E*| dynamic moduli were assembled and imported into the MEPDG for temperature scenarios of 66, 113, 147, and 165 °F (19, 45, 64, and 74 °C). Then, the moduli from the first month of the MEPDG was selected and input to the standalone application for a fixed modulus that does not age and stiffen with time. These moduli for the HMA layers are summarized in table 52. Although the temperature was designed to be fixed with depth and time using modified input files from the EICM, moduli from the MEPDG did vary slightly with depth. The moduli reported in table 52 are the average of all sublayers in the HMA. The typical variation of moduli from top to bottom was 6 percent for the 4-inch (100‑mm) lanes at 147 and 165 °F (64 and 74 °C), 16 percent for the 4-inch (100-mm) lanes at 66 °F (19 °C), 9 percent for the 5.8-inch (150-mm) lanes at 147 °F (64 °C), 17 percent for the 5.8-inch (150-mm) lanes at 113 °F (45 °C), and 17 percent for the 5.8-inch (150-mm) lanes at 66 °F (19 °C).

Table 52. First-month modulus from MEPDG used in standalone program.

Lane

Site

Temperature (°C)

As-Built HMA Modulus (psi)

As-Designed HMA Modulus (psi)

2; PG70-22

1

64

25,294

42,099

2

74

14,596

22,379

3

19

1,019,332

1,235,002

3; air blown

1

64

35,633

54,837

2

74

20,229

30,687

3

19

811,735

947,703

4; SBS-LG

1

64

23,070

26,881

2

74

14,315

17,035

3

19

716,739

711,956

5; CR-TB

1

64

50,716

37,358

2

74

27,040

21,552

3

19

855,027

718,840

6; terpolymer

1

64

17,579

21,721

2

74

11,764

14,822

3

19

647,964

654,057

8; PG70-22

1

64

26,684

41,723

2

45

117,266

176,840

3

19

1,078,588

1,225,775

9; SBS 64-40

1

64

17,643

16,314

3

19

371,394

298,872

10; air blown

1

64

40,874

54,427

2

45

146,546

182,274

3

19

925,573

941,817

11; SBS-LG

1

64

25,200

26,682

2

45

91,866

88,306

3

19

780,127

703,421

12; terpolymer

1

64

20,107

21,541

2

45

74,043

67,062

3

19

744,576

640,028

°F = 1.8(°C) + 32
1 psi = 6.89 kPa

It was recognized that high-temperature ALF rutting at 147 and 165 °F (64 and 74 °C) was predicted with MEPDG models using dynamic modulus master curves that were generated from physical test data where the highest temperature was only 136 °F (58 °C). This created some concern that material properties were being extrapolated beyond the range of what was physically measured. However, time-temperature superposition was used to create the dynamic modulus master curves. Time-temperature superposition allows temperature effects and rate effects (time, frequency) to be interchanged. In other words, the dynamic modulus at a cooler temperature and given frequency is equivalent to the dynamic modulus at a warmer temperature and higher frequency; temperature and rate effects are inversely related. Table 53 shows an example of the dynamic modulus master curve data from the high-temperature portion of the control lab-produced mixture. The non-reduced frequencies and corresponding temperatures are provided in the left portion of the table and correspond to the individual reduced frequencies (at a 66 °F (19 °C) reference temperature) and dynamic modulus counterpart in the right side of the table. The nearest equivalent dynamic modulus from temperatures and frequencies of 147 °F (64 °C) at 10 Hz and 165 °F (74 °C) at 10 Hz are equivalently the dynamic modulus at 136 °F (58 °C) at frequencies near 5 and 1 Hz, respectively. The equivalent computed frequencies at 136 °F (58 °C) are 3.4 and 0.65 Hz for identical modulus at 147 °F (64 °C) at 10 Hz and 165 °F (74 °C), respectively.

Table 53. Example of equivalent |E*| temperatures and frequencies using time-temperature superposition.

Temperature
(°C)

Frequency (Hz)

Reduced Frequency (Hz)

|E*|

MPa

psi

58

20

3.09E-03

492

71,323

58

10

1.55E-03

360

52,157

64

20

1.06E-03

303

43,906

58

5

7.73E-04

263

38,179

64

10

5.28E-04

222

32,205

64

5

2.64E-04

164

23,774

74

20

2.07E-04

148

21,424

58

1

1.55E-04

131

18,932

74

10

1.04E-04

111

16,030

58

0.5

7.73E-05

98

14,240

64

1

5.28E-05

84

12,244

74

5

5.18E-05

84

12,153

64

0.5

2.64E-05

65

9,420

58

0.1

1.55E-05

54

7,784

74

1

1.04E-05

47

6,791

64

0.1

5.28E-06

38

5,474

74*

0.5*

5.18E-06*

38*

5,442*

74*

0.1*

1.04E-06*

24*

3,489*

°F = 1.8(°C) + 32

*Extrapolation beyond laboratory measured conditions.

Quantifying Cracking

Predicted cracking quantified by the MEPDG and measured under the ALF are different in nature. ALF measures percent cracking within the boundary of the area loaded by the single tire. With wheel wander, the dimensions of the ALF loaded area were 3.7 by 33 ft (1.13 by 10.05 m), equaling 122.6 ft2 (11.4 m2). The percent ALF cracking is calculated using figure 86.

ALF cracking equals 100 percent times Area subscript cracked divided by Areas subscript total which is equal to 100 percent times a given x amount of cracked area divided by 1.13 meters times 10.05 meters, which is equal to 100 percent times a given x amount of cracked area divided by 11.4 square meters.

Figure 86. Equation. ALF cracking.

The equivalent MEPDG cracking for the same amount of ALF cracking can be calculated assuming two loaded areas within a standard lane width of 12 ft (3.66 m), as shown in figure 87.

Equivalent MEPDG cracking equals 100 percent times 2 times a given x amount of cracked area divided by 3.66 meters times 10.05 meters, which is equal to 100 percent times a given x amount of cracking divided by 36.8 square meters.

Figure 87. Equation. Equivalent MEPDG cracking.

A factor of 0.31 is used to adjust the ALF cracking to what is comparable with MEPDG. It is calculated by dividing the two reference areas, 122.6 ft2 (11.4 m2) divided by 120.7 ft2 (36.8 m2).

Primary Strain Response of Pavements

Strain gauges were installed in one of the four sites of each lane. Figure 88 is schematic of the strain gauge layout and orientation. An evaluation of the measured strain gauge response against that predicted by the linearly elastic mechanistic material properties provides a qualitative assessment of those assigned linear elastic properties. It must be recognized that the strain gauge measured responses cannot be taken as true since the presence of the gauge embedded within the HMA layer and circumstances of installation and orientation can affect the response. However, strain gauges can provide reality checks on the layered elastic mechanistic pavement models and chosen material property inputs.

This illustration shows a plan-view schematic layout of an accelerated loading facility (ALF) test site. A rectangular section delineates the area of wheel wander and crack mapping with stationing marked along the sides. The location of three longitudinal and two horizontal asphalt strain gauges are illustrated.
1 ft = 0.305 m
1 inch = 25.4 mm

Figure 88. Illustration. Layout of an ALF test site with strain gauges.

Strain gauge responses were measured before any fatigue or rutting loading began. HMA tensile strains were measured under a variety of temperatures, wheel loads, and transverse wheel offsets. The series of tests that was closest to the ALF fatigue loading conditions and corresponding MEPDG standalone output (66 °F (19 °C) and 16,000 lbf (70 kN)) were strain measurements controlled at 66 °F (19 °C) at 14,500 lbf (64 kN). Figure 89 and table 54 compare the measured HMA tensile strains against layered elastic MEPDG predicted strains from the as-built scenario where the HMA thickness and unbound layer moduli as well as HMA moduli were adjusted at each lane and site. When multiple transverse or longitudinal strain gauges survived construction, the average strain is provided along with the standard deviation, represented by error bars. The line of equality is included in the figure, and all data points lie above the line, which indicates the predicted strains using the larger 16,000-lbf (70-kN) wheel loads are still smaller than the measured strains from the 14,500-lb (64 kN) wheel load. However, there is a strong ranking and relationship. For the relationship between predicted strains and measured transverse strains, the slope is positive with a value of 1.41 and the Kendall’s tau score is +0.6. The significance of Kendall’s tau score and ANOVA significance of the regression are very good at 99 and 97 percent, respectively. The R2 value is 0.47. For the relationship between predicted strains and measured longitudinal strains, the slope is positive with a value of 1.27, and Kendall’s tau score is also +0.6. The significance of Kendall’s tau score and ANOVA significance of the regression are very good, both at 99 percent. The R2 value is 0.65.

Table 54. Measured and predicted HMA tensile strains.

Lane

14.5 kip Measured HMA Tensile Strain (με)

16 kip Predicted MEPDG Standalone (με)

Transverse

Longitudinal

2, PG70-22

496

493

397

3, Air blown

689

622

578

4, SBS-LG

976

862

633

5, CR-TB

1,524

1,080

549

6, Terpolymer

927

1,038

624

7, Fiber

539

550

8, PG70-22

476

370

268

9, SBS 64-40

1,042

926

614

10, Air blown

488

426

271

11, SBS-LG

600

565

272

12, Terpolymer

690

671

304

1 kip = 4.45 kN

— Indicates the simulation was not performed for lane 7.

This plot shows measured hot mix asphalt (HMA) tensile strain on the y-axis and predicted HMA tensile strain on the x-axis. A line of equality with a one-to-one slope is provided. Two series of data points are provided from the longitudinal and transverse strain gauges, showing the same trend above the line of equality. Each data point is for a particular ALF lane, and error bars represent the standard deviation when multiple replicate data points were available.

Figure 89. Graph. Measured HMA tensile strain versus predicted HMA tensile strain.

The strain gauges were monitored during fatigue loading to detect the accumulation of fatigue damage that ideally would be registered by an increase in strain to reflect reduction in modulus. A typical example of the strain gauge response is provided from lane 3 (air blown) in figure 90, which shows a small increase in the strain during the early loading, as would be expected. Counterintuitively, the strain decreases after this point. The interpretation of this response is that these particular types of embedded strain gauges are only reliable to assess the initial undamaged properties of HMA layers. It is hypothesized that as the HMA becomes damaged the gauge loses strength and the ability to remain anchored and embedded in the material and registers smaller strains than are actually occurring.

This graph shows measured tensile strain on the left y-axis versus the number of accelerated loading facility (ALF) passes on the x-axis. The right y-axis shows the percent of the loaded area cracked due to fatigue. The figure illustrates a typical observation in the different ALF test lanes. Three series of longitudinal strain and two series of transverse strain show a rapid increase in strain early, up to about 8,000 cycles, followed by a gradual reduction in strain to 90,000 cycles. Cracking increases rapidly after about 8,000 cycles.

Figure 90. Graph. Lane 3 measured tensile strain versus number of ALF passes.

Measured Reduction of Modulus

Nondestructive pavement evaluation was conducted during fatigue loading of lane 8 (PG70-22) and lane 10 (air blown). A portable seismic pavement analyzer (PSPA) was used to assess changes in the HMA modulus as fatigue loading accumulated. Theoretically, small microscopic cracks, or microcracks, develop and grow in the material, thereby reducing the modulus to the point where larger macrocracks localize and then propagate through the material and manifest as alligator fatigue cracks observed on the surface. Measurements were made in four locations in the center of the wheel path in both the longitudinal and transverse directions.

Results from lanes 8 and 10 are shown in figure 91 and figure 92, respectively, with the measured seismic modulus on the left axis and the measured ALF cracking on the right axis. There is variability in the measured modulus, but a trend shows the modulus reducing with more and more passes until, ultimately, fatigue cracks can be observed on the surface. There appears to be no difference between modulus measured in the longitudinal and transverse directions. After fatigue cracks reach the surface, the PSPA-measured modulus becomes very erratic, especially in the large amount of cracking observed in lane 10.

This graph shows in-situ measured hot mix asphalt (HMA) modulus with seismic analysis on the left y-axis versus the number of accelerated load facility (ALF) passes on the x-axis for lane 8 (performance grade 70-22). The right y-axis shows the percent of the loaded area cracked due to fatigue. A series of longitudinal seismic modulus and transverse seismic modulus are in agreement with each other, both showing a gradual reduction with ALF passes while the pavement remains uncracked. Only later, when significant modulus has been lost, does the cracked area begin to increase.
1 GPa = 145,000 psi

Figure 91. Graph. Lane 8 in situ measured HMA modulus with seismic analysis versus number of ALF passes.

 

This graph shows in-situ measured hot mix asphalt (HMA) modulus with seismic analysis on the left y-axis versus the number of accelerated load facility (ALF) passes on the x-axis for lane 10 (air blown). The right y-axis shows the percent of the loaded area cracked due to fatigue. A series of longitudinal seismic modulus and transverse seismic modulus are in agreement with each other, both showing a gradual reduction with ALF passes while the pavement remains uncracked. Only later, when significant modulus has been lost, does the cracked area begin to increase. The seismic modulus becomes erratic when the cracked area becomes large.
1 GPa = 145,000 psi

Figure 92. Graph. Lane 10 in situ measured HMA modulus with seismic analysis versus number of ALF passes.

Evaluation of Damage in Uncracked Lanes 11 and 12

PSPA tests were conducted in the loaded area of site 3 and the unloaded area of site 4 to assess whether reduction in modulus could confirm the estimated ranking of the fatigue cracking performance for lanes 11 and 12, as described in chapter 3. The average and standard deviation of PSPA modulus was measured at four locations along the wheel path, each repeated three times. The moduli are plotted in figure 93 and figure 94. The moduli from unloaded site 4 correspond to zero ALF passes while the moduli from loaded site 3 correspond to the maximum passes each lane received: 400,000 passes for lane 12 and 673,000 passes for lane 11. A statistical analysis indicates there are no significant differences between the moduli measured at zero passes and 673,000 passes for lane 11 (SBS-LG) while lane 12 (terpolymer) showed the modulus at 400,000 passes was statistically smaller than at zero passes. This suggests that the lane 11 SBS-LG is more resistant to fatigue damage than the lane 12 terpolymer.

 This graph shows measured seismic modulus on the y-axis and number of accelerated load facility (ALF) passes on the x-axis for lane 11 (linear grafted styrene-butadiene-styrene). Two data points link the modulus measured at zero passes and at greater than 600,000 passes. The trend is flat, with error bars at each data point representing a standard deviation of measured seismic modulus.
1 ksi = 6.89 kPa

Figure 93. Graph. Lane 11 measured seismic modulus versus number of ALF passes.

 

This graph shows measured seismic modulus on the y-axis and number of accelerated load facility (ALF) passes on the x-axis for lane 12 (terpolymer). Two data points link the modulus measured at zero passes and at greater than 400,000 passes. The trend is downward, with error bars at each data point representing a standard deviation of measured seismic modulus.
1 ksi = 6.89 kPa

Figure 94. Graph. Lane 12 measured seismic modulus versus number of ALF passes.

PREDICTED PERFORMANCE FROM MEPDG STANDALONE PROGRAM

Influence of Construction Variability on Rutting

All predicted performance from the MEPDG standalone in this report is at a 50 percent reliability level rather than the customary 98 percent reliability. The magnitude of rutting predicted by the MEPDG standalone program was significantly larger than what was produced by ALF. Rut depths on the order of 2 to 5 inches (51 to 127 mm) are typical, as seen in the example given in figure 95 for the 147 °F (64 °C) as-built simulations. However, it is unfair to compare the predictive competencies of the nationally calibrated MEPDG against the extreme heavy wheel loads and zero wander or channeled rutting. The empirical rutting distress model used by the MEPDG is given in figure 96. It is likely the temperature term, , and high-temperature input of the ALF conditions are major contributors to high amounts of calculated rutting.

This graph shows predicted curves of rutting from the Mechanistic-Empirical Pavement Design Guide (MEPDG) on the y- axis and number of accelerated load facility (ALF) passes on the x-axis at 147 °F (64 °C). Five series of rutting predicted from 4 inch (100-mm) lanes are shown as solid lines, and five series of rutting predicted from 5.8-inch (150 mm) lanes are shown as dashed lines. The families of curves are interspersed and vary between 2 and 5 inches (51 and 127 mm) at 40,000 passes.
1 inch = 25.4 mm

Figure 95. Graph. Predicted curves of rutting from the MEPDG versus number of ALF passes at 147 °F (64 °C).

 

Epsilon subscript p divided by epsilon subscript r equals k subscript z times the quantity beta subscript r1 times the quantity 10 raised to the power k subscript 1 times the quantity T raised to the power k subscript 2 times beta times r subscript 2 times the quantity N raised to the power k subscript 3 times beta times r subscript 3.

Figure 96. Equation. Empirical rutting distress model used by MEPDG.

Where:

Epsilon subscript p divided by epsilon subscript r = Ratio of vertical plastic strain to vertical recoverable elastic strain.

kZ= Depth correction factor.

T = Temperature, °F.

N = Number of load passes.

Several other noteworthy observations are in regards to the predicted and measured rutting. The ranking of the MEPDG predicted rutting does not change regardless of the amount of cycles. The predicted curves do not crisscross, as observed in some of the measured data given in figure 24 through figure 27, figure 46, and figure 47. This is due to the fact that a single curve fit empirical power law (figure 96) is adjusted up or down by the dynamic modulus |E*| input via the recoverable strain term, . In other words, the correlation of MEPDG-predicted rutting is directly linked to the stiffness of the material inputs. This is not a criticism of the MEPDG, just a consequence of the nationally calibrated coefficients. However, this was justified in NCHRP 9‑19, which showed that, generally, stiffer pavement ruts less.(64)

Table 55 through table 58 show the predicted rutting from the MEPDG standalone for three scenarios each at 147, 165, 113, and 66 °F (64, 74, 45, and 19 °C). As previously described, the as-built scenario was the most faithful to the conditions of the ALF and considered the variation in HMA thickness, differences in HMA stiffness and density, and variation in unbound base and subgrade stiffness. The as-built with average unbound layer modulus scenario fixed the modulus of the unbound base and subgrade for all lanes at the average value. The as-designed scenario fixed the unbound layer moduli of all lanes, used the ideal 4- or 5.8-inch (100- or 150-mm) thickness of the HMA layers, and used the lab-produced HMA |E*| at a single fixed air void content of 7 percent.

Table 55. Predicted 147 °F (64 °C) (40,600 passes) rutting from MEPDG standalone.

Location

As-Built

As-Built with Average Unbound Layer Modulus

As-Designed

MEPDG

Standalone

Rut Depth (inches)

Ranking

MEPDG

Standalone

Rut Depth (inches)

Ranking

MEPDG

Standalone

Rut Depth (inches)

Ranking

Lane 5, CR-TB, 100 mm

1.87

1

1.84

1

2.50

5

Lane 10, air blown, 150 mm

2.06

2

2.06

2

1.40

1

Lane 3, air blown, 100 mm

2.60

3

2.70

3

1.67

2

Lane 8, PG70-22, 150 mm

3.43

4

3.47

4

2.00

3

Lane 11, SBS-LG, 150 mm

3.80

5

3.60

5

3.40

6

Lane 2, PG70-22, 100 mm

3.96

6

3.88

6

2.20

4

Lane 4, SBS-LG, 100 mm

4.20

7

4.26

7

3.60

7

Lane 12, terpolymer, 150 mm

5.00

8

4.80

8

4.40

8

Lane 9, SBS 64-40, 150 mm

5.50

9

5.65

9

6.08

10

Lane 6, terpolymer, 100 mm

5.70

10

5.86

10

4.60

9

1 mm = 0.039 inches

Table 56. Predicted 166 °F (74 °C) (40,600 passes) rutting from MEPDG standalone.

Location

As-Built

As-Built with Average Unbound Layer Modulus

As-Designed

MEPDG

Standalone

Rut Depth (inches)

Ranking

MEPDG

Standalone

Rut Depth (inches)

Ranking

MEPDG

Standalone

Rut Depth (inches)

Ranking

Lane 5, CR-TB, 100 mm

4.20

1

4.30

1

5.50

3

Lane 3, air blown, 100 mm

5.75

2

5.90

2

3.70

1

Lane 4, SBS-LG, 100 mm

8.30

3

8.50

3

7.15

4

Lane 2, PG70-22, 100 mm

8.40

4

8.60

4

5.30

2

Lane 6, terpolymer, 100 mm

10.70

5

10.90

5

8.30

5

1 mm = 0.039 inches

Table 57. Predicted 113 °F (45 °C) (302,064 passes) rutting from MEPDG standalone.

Location

As-Built

As-Built with Average Unbound Layer Modulus

As-Designed

MEPDG

Standalone

Rut Depth (inches)

Ranking

MEPDG

Standalone

Rut Depth (inches)

Ranking

MEPDG

Standalone

Rut Depth (inches)

Ranking

Lane 10, air blown, 150 mm

0.74

1

0.73

1

0.55

1

Lane 8, PG70-22, 150 mm

0.97

2

0.97

2

0.58

2

Lane 11, SBS-LG, 150 mm

1.38

3

1.30

3

1.37

3

Lane 12, terpolymer, 150 mm

1.85

4

1.77

4

1.98

4

1 mm = 0.039 inches

Table 58. Predicted 66 °F (19 °C) (302,064 passes) rutting from MEPDG standalone.

Location

As-Built

As-Built with Average Unbound Layer Modulus

As-Designed

MEPDG

Standalone

Rut Depth (inches)

Ranking

MEPDG

Standalone

Rut Depth (inches)

Ranking

MEPDG

Standalone

Rut Depth (inches)

Ranking

Lane 8, PG70-22, 150 mm

0.040

1

0.040

1

0.030

1

Lane 10, air blown, 150 mm

0.040

2

0.040

2

0.040

2

Lane 11, SBS-LG, 150 mm

0.040

3

0.045

3

0.050

3

Lane 12, terpolymer, 150 mm

0.050

4

0.060

4

0.060

4

Lane 2, PG70-22, 100 mm

0.100

5

0.100

5

0.110

5

Lane 9, SBS 64-40, 150 mm

0.110

6

0.100

6

0.110

6

Lane 5, CR-TB, 100 mm

0.160

7

0.140

7

0.150

8

Lane 3, air blown, 100 mm

0.170

8

0.150

8

0.130

7

Lane 6, terpolymer, 100 mm

0.170

9

0.150

9

0.170

10

Lane 4, SBS-LG, 100 mm

0.200

10

0.180

10

0.160

9

1 mm = 0.039 inches

A comparison between the relative ranking of the predicted rutting from the as-built and as-built with average unbound layer modulus scenarios provides insight as to the likelihood that the variation in unbound layer moduli influenced the measured rutting. This is important because the strength of binder properties and mixture properties to account for rutting performance is assessed by comparisons against the ALF performance. The identical rankings between these two scenarios for all rutting temperatures indicate that the variation in unbound layer moduli did not negatively impact the measured rutting performance of the ALF.

A comparison of the relative rankings between the predicted rutting from the as-built with average unbound layer modulus and as-designed scenarios provides insight as to the likelihood that the variation in HMA stiffness and density influenced the measured rutting. Again, this is important because the strength of binder properties and mixture properties to account for rutting performance is assessed by comparisons against the ALF performance. There is mild variation in the relative ranking between these scenarios, quantified with the Kendall’s tau score and significance. The Kendall’s tau scores are +0.73, +0.4, +1.0, and +0.91 for 147, 165, 113, and 66 °F (64, 74, 45, and 19 °C), respectively. The statistical significance was 99.9, 76, 100, and 91 percent, respectively. Overall, this suggests that the variation in HMA density and stiffness had a weak effect on the measured performance, except at 165 °F (74 °C). This could in part be because there were fewer data points at 165 °F (74 °C). Nonetheless, it must be recognized that the predicted rutting was impractically large for the three highest temperatures without lateral wheel wander.

Influence of Construction Variability on Fatigue Cracking

Predicted bottom-up fatigue cracking from the MEPDG standalone program is shown in figure 97 through figure 99 for the three different scenarios. Ranked predicted fatigue cracking is shown in table 58. The shape of the predicted cracking curve is smooth and continuously increasing from the beginning at zero passes. This is a marked difference from the measured pattern of cracking where no fatigue cracks are registered until they reach the surface and then increase in an almost linear fashion. Again, this is merely an observation, not a criticism of the nationally calibrated MEPDG. Unlike predicted rutting, the fatigue cracking curves from the different mixtures and lanes can crisscross.

This graph shows percent cracking on the y-axis and number of accelerated load facility passes on the x-axis and uses fatigue cracking predicted by the Mechanistic-Empirical Pavement Design Guide (MEPDG) standalone program. The group of curves from the 5.8 inch (150-mm) lanes is much smaller than the group of curves from the 4 inch (100-mm) lanes.

Figure 97. Graph. Percent fatigue cracking predicted from MEPDG standalone program for the as-built scenario.

 

This graph shows percent cracking on the y-axis and number of accelerated load facility passes on the x-axis and uses fatigue cracking predicted by the Mechanistic-Empirical Pavement Design Guide (MEPDG) standalone program. The group of curves from the 5.8-inch (150-mm) lanes is much smaller than the group of curves from the 4-inch (100-mm) lanes.

Figure 98. Graph. Percent fatigue cracking predicted from MEPDG standalone program for the as-built with average unbound layer modulus scenario.

 

This graph shows percent cracking on the y-axis and number of accelerated load facility passes on the x-axis and uses fatigue cracking predicted by the Mechanistic-Empirical Pavement Design Guide (MEPDG) standalone program. The group of curves from the 5.8-inch (150-mm) lanes is smaller than the group of curves from the 4-inch (100-mm) lanes, with limited interspersion between.

Figure 99. Graph. Percent fatigue cracking predicted from MEPDG standalone program for the as-designed scenario.

Table 59. Predicted 66 °F (19 °C) (302,064 passes) fatigue cracking from MEPDG standalone.

Location

As-Built

As-Built with Average Unbound Layer Modulus

As-Designed

MEPDG Standalone Cracking (percent)

Ranking

MEPDG Standalone Cracking (percent)

Ranking

MEPDG Standalone Cracking (percent)

Ranking

Lane 11, SBS-LG, 150 mm

1.30

1

2.74

3

6.57

3

Lane 10, air blown, 150 mm

1.53

2

2.11

2

4.20

2

Lane 8, PG70-22, 150 mm

1.56

3

1.50

1

2.76

1

Lane 12, terpolymer, 150 mm

1.68

4

3.25

4

7.70

4

Lane 9, SBS 64-40, 150 mm

6.64

5

4.90

5

18.60

6

Lane 2, PG70-22, 100 mm

25.50

6

22.50

8

14.50

5

Lane 3, air blown, 100 mm

30.40

7

17.70

6

20.50

7

Lane 5, CR-TB, 100 mm

30.90

8

19.30

7

27.20

8

Lane 6, terpolymer, 100 mm

42.20

9

26.30

9

30.40

10

Lane 4, SBS-LG, 100 mm

42.50

10

31.50

10

27.60

9

1 mm = 0.039 inches

The same analysis used for the predicted rut depth ranking of the three scenarios was conducted on the predicted fatigue from the MEPDG standalone. A comparison of the relative rankings between the predicted fatigue from the as-built and as-built with average unbound layer modulus scenarios provides insight as to the likelihood that the variation in unbound layer moduli influenced the measured cracking. Again, this is important because the strength of binder and mixture properties to account for fatigue performance is assessed by comparisons against the ALF performance. Kendall’s tau score is +0.78, and the statistical significance is over 99.9 percent. This suggests the variation in unbound layer stiffness had a weak effect on the measured performance.

The comparison of fatigue cracking between the as-built with average unbound layer modulus and as-designed scenarios was repeated to evaluate the impacts that variations in HMA thickness and stiffness (density) may have had on the measured fatigue. Kendall’s tau score is +0.82, and the statistical significance is over 99.9 percent. This suggests the variation in unbound layer stiffness had a weak effect on the measured performance.

The above comparisons combined the data from both the 4- and 5.8-inch (100- and 150-mm) sections. However, the predicted fatigue cracking in the thicker, 5.8-inch (150-mm) sections was always much less than the thinner, 4-inch (100-mm) sections, and this could numerically mask an important comparison. Thus, the variation in ranking within the 4- and 5.8-inch (100- and 150-mm) sections was analyzed separately, which utilized a more direct comparison of ranking between the as-built and as-designed 4-inch (100-mm) sections and between the as-built and as-designed 5.8-inch (150-mm) sections. For the 4-inch (100-mm) sections, Kendall’s tau score is +0.8, and the statistical significance is 95.8 percent. For the 150-mm sections, Kendall’s tau score is +0.4, and the statistical significance is 75.8 percent. Overall, this indicates that construction variation had less of an influence on performance than unbound layer stiffness and a minor influence on the measured fatigue cracking. When these values are considered in light of the actual, wider differences in measured fatigue cracking under the ALF, the ranking of the sections may not have changed significantly if constructed at the ideal density and thickness.

Assessment of MEPDG Predictive Capability

As previously stated, it is unfair to expect a high degree of predictive accuracy of the MEPDG for the ALF test sections because the MEPDG uses a national calibration, the national calibration is based almost entirely on HMA mixtures having unmodified asphalt binders, and the heavy wheel loads and intense number of cycles challenge the calibration that took place under more natural traffic. Only the relative order of measured versus predicted distress is compared rather than absolute predictive capability, and the analysis should not be taken as a weakness or negative departure from the intended use of the MEPDG. As the name suggests, the MEPDG is meant to provide design guidance to practitioners. As implementation proceeds, the primary utilization of the MEPDG will be to understand relative changes in performance due to the consequences of changing a particular layer’s material, stiffness, thickness, etc.

Rutting

MEPDG standalone rutting was compared against measured rutting at all temperatures, and all 4- and 5.8-inch (100- and 150-mm) data points were combined. The rut depths at a fixed number of load cycles under the ALF and calculated by the MEPDG standalone were directly compared against one another. The number of cycles at which the measured and predicted rutting were compared depended on the temperature of the test and thickness of the HMA layer. At 66 °F (19 °C), where lateral wheel wander was utilized for fatigue cracking, the rutting was compared at 100,000 cycles for the 4-inch (100-mm) lanes and at 300,000 cycles for the 5.8-inch (150-mm) lanes. At 113, 147, and 165 °F (45, 64, and 74 °C), the rutting was compared at 200,000, 25,000, and 10,000 cycles, respectively. The average measured rutting for these comparisons was 0.4 inches (9.8 mm). The minimum and maximum measured rutting were 0.1 and 0.7 inches (3.3 and 17.3 mm).

The as-built with average unbound layers scenario was not considered. Two extreme simulation scenarios were considered, as-built and as-designed, and are shown in figure 100 and figure 101, respectively. It is clear that the rutting from higher temperatures without lateral wheel wander was over-predicted, while the rutting from the 66 °F (19 °C) intermediate temperature with lateral wheel wander was under-predicted. As noted previously, large impractical rutting does not necessarily detract from the MEPDG because of the calibration and ALF conditions. Regardless of which simulation scenario is considered, the MEPDG does account for different amounts of rutting at different temperatures and cycles. There is a positive proportional relationship between measured and predicted rutting. Kendall’s tau parameters from as-built and as-designed inputs were +0.50 and +0.46, respectively. The significance of the regression from as-built and as-designed inputs were 99.99 percent (a p-value of 0.001 percent) and 99.9 percent (a p-value of 0.010 percent), respectively. R from as-built and as-designed inputs were +0.71 and +0.66, respectively. R2 from as-built and as-designed scenarios were 0.51 and 0.43, respectively.

This graph shows measured accelerated load facility (ALF) rutting on the y-axis and Mechanistic-Empirical Pavement Design Guide (MEPDG) standalone-predicted rutting on the x-axis for the as-built scenario. Five groups of data points with different are shown, representing 66 °F (19 °C) rutting in 4-inch (100-mm) lanes, 66 °F (19 °C) rutting in
1 mm = 0.039 inches

Figure 100. Graph. Measured ALF rutting versus MEPDG standalone-predicted rutting for the as-built scenario.

 

This graph shows measured accelerated load facility (ALF) rutting on the y-axis and Mechanistic-Empirical Pavement Design Guide (MEPDG) standalone-predicted rutting on the x-axis for the as-designed scenario. Five groups of data points are
1 mm = 0.039 inches

Figure 101. Graph. Measured ALF rutting versus MEPDG standalone-predicted rutting for the as-designed scenario.

 

The measured and predicted rut data at each temperature were assessed without the presence of other temperature data to artificially improve the statistics. Table 60 provides the different techniques’ statistical measures, where the general trend is that the as-built simulations provide slightly better results than the as-designed scenarios. However, regardless of as-built or as-designed, the individual temperature data indicate poor ranking, where Kendall’s tau is quite low but with a very mediocre significance and the data at 165 °F (74 °C) are essentially too few and too poor to draw any meaningful conclusions. For the same reasons, the three data points from 113 °F (45 °C) were omitted.

Table 60. Statistical analysis of measured and predicted rutting at different temperatures

Scenario

Temperature
(°C)

Slope

Regression Significance (1 – p-value)

Kendall’s Tau

Kendall’s Tau Significance (percent)

R

R2

As-built inputs

19*

+0.71

47.1

+0.07

56.9

+0.22

0.05

64*

+0.05

94.6

+0.34

89.0

+0.59

0.35

74**

+0.08

75.8

+0.60

88.3

+0.64

0.41

As-designed inputs

19*

+0.15

8.6

+0.11

63.0

+0.04

0.00

64*

+0.04

87.7

+0.24

77.9

+0.49

0.24

74**

+0.04

28.7

0

40.8

+0.22

0.05

*n = 10.
**n = 5.

In summary, the MEPDG model is valid and able to account for differences in rutting between different temperatures and cycles. The statistics are very weak but suggest that the MEPDG is sensitive to the type of inputs being as-designed or as-built. The statistics from the as-built predictions are slightly stronger than the statistics from the as-designed scenarios. However, the analysis indicates the predictive capabilities of the current national calibration and corresponding mechanistic-empirical models are poor and cannot distinguish between the performance due only to asphalt binder at particular temperatures and load levels.

Fatigue Cracking

Predicted fatigue cracking distress is computed from two models within the MEPDG. First, a mechanistic-empirical model predicts the number of cycles to fatigue cracking failure (NF) that is a function of the modulus of the HMA at a particular time and the tensile strain from a particular axle load (see figure 102).(1) The amount of fatigue cracking distress (percent cracked area) reported by the MEPDG is then computed using an empirical model and Miner’s Law to consider accumulated damage from the NF equation for different combinations of traffic and environmental conditions.

N subscript f equals 0.00432 times C times the quantity k subscript 1 times the quantity open parenthesis one divided by epsilon subscript T closed parenthesis raised to the power k subscript2 times the quantity open parenthesis one divided by E closed parenthesis raised to the power k subscript 3.

Figure 102. Equation. Cycles to fatigue cracking failure.

Where:

N subscript f equals 0.00432 times C times the quantity k subscript 1 times the quantity open parenthesis one divided by epsilon subscript T closed parenthesis raised to the power k subscript2 times the quantity open parenthesis one divided by E closed parenthesis raised to the power k subscript 3.

εT = HMA tensile strain (mm/mm or inch/inch).

E = HMA modulus (psi).

Va = Air void content (percent).

Vb = Effective volume of binder (percent).

hAC = Thickness of HMA (inches).

Two methods were used to assess the predictive capability of the MEPDG. Standard output of percent cracked area from the MEPDG standalone was directly compared to the cracked area from ALF. The number of ALF cycles to achieve 25 percent cracked area, which is equivalent to 7.75 percent MEPDG cracked area, was compared to the predicted number of cycles to 7.75 percent cracked area. Also, the internal mechanistic-empirical NF model was evaluated by comparing the number of ALF cycles to initiate surface cracks and calculated number of cycles to failure computed from the known strains in table 54 and known |E*| dynamic moduli in table 52.

Measured fatigue cracking data in chapter 3 clearly show that many more cycles were required to induce fatigue cracking in the thicker 5.8-inch (150-mm) lanes than in the thinner 4-inch (100‑mm) lanes and that additional analyses using extrapolation and interpolation were needed to develop a complete set of ranked performance to common criteria. The comparison between measured and predicted number of cycles to 7.75 percent cracked area is shown in arithmetic scale in figure 103 and in log-log scale in figure 104 to highlight the fatigue cracking data points from the 4-inch (100-mm) sections. Unlike predicted rutting, the measured and fatigue cracking data points are scattered above and below the line of equality for both the 4- and 5.8-inch (100- and 150-mm) sections. The predicted fatigue cracking is still qualitatively inaccurate but not as impractical as the predicted rutting. Also unlike rutting, the trends become negative instead of proportionally positive when a smaller data subset is evaluated. Smaller sets of temperature data were evaluated in rutting, and smaller sets of thickness were evaluated in fatigue.

This graph shows three families of data points for the measured versus as-built, as-designed, and as-built with average unbound modulus scenarios. A line of equality is included for reference where there is significant degree of scatter. Predicted cycles to 7.75 percent cracked area from the Mechanistic-Empirical Pavement Design Guide (MEPDG) is shown on the y-axis in arithmetic scale, and corresponding measured number of cycles is shown on the x-axis in arithmetic scale.

Figure 103. Graph. Predicted cycles to 7.75 percent cracked area from MEPDG in arithmetic scale versus measured number of cycles.

 

This graph shows three families of data points for the measured versus as-built, as-designed, and as-built with average unbound modulus scenarios. A line of equality is included for reference where there is significant degree of scatter. Predicted cycles to 7.75 percent cracked area from the Mechanistic-Empirical Pavement Design Guide (MEPDG) is shown on the y-axis in log scale, and corresponding measured number of cycles is shown on the x-axis in log scale. Data points from the 4-inch (100-mm) lanes are clustered apart from the data points from the 5.8-inch (150-mm) lanes; both sets are without a proportionally positive trend.

Figure 104. Graph. Predicted cycles to 7.75 percent cracked area from MEPDG in log scale versus measured number of cycles.

Figure 105 shows the measured number of cycles to achieve surface cracking plotted against the mechanistic-empirical number of cycles to failure computed from figure 102. Although unexpected, the data are distributed on either side of the line of equality. The overall relationship is proportionally positive. The data points from the 4- and 5.8-inch (100- and 150-mm)-thick pavements are not identified in the figure, but data from the 4-inch (100-mm) lanes are in the lower left side, similar to figure 104.

This graph shows predicted cycles to failure from the Mechanistic-Empirical Pavement Design Guide (MEPDG) equation on the y-axis in arithmetic scale and measured number of cycles to surface crack initiation on the x-axis in arithmetic scale. Two groups of data points illustrate similar trends for as-built modulus and as-designed modulus scenarios. A line of equality is included for reference with notable, yet symmetrical, scatter about the line of equality and with a proportional positive trend.

Figure 105. Graph. Predicted cycles to failure from MEPDG equation in arithmetic scale versus measured number of cycles to surface crack initiation.

Table 61 contains the statistical measures for the various comparisons of ALF performance with MEPDG cracking output and the internal mechanistic-empirical model for number of cycles to failure. The trends are essentially very weak but go in the correct direction when data from both 4- and 5.8-inch (100- and 150–mm) lanes are considered together. The regression slope, Kendall’s tau, and correlation coefficient are all positive, but the statistical significance is smaller than conventionally desired, such as above 90 percent. The trends appear to get weaker as the simulation scenario diverges from as-built to as-designed. However, when only the 4-inch (100-mm) lanes are considered (where the dataset is more complete than the 5.8-inch (150-mm) lanes), the comparisons are worse in both cases of cracking output and number of cycles to failure. Although the strength of the relationships between measured and predicted is stronger, the direction of the relationship is now inverse and not reasonable.

Table 61. Statistical analysis of measured and predicted fatigue cracking.

Source

Scenario

Slope

Regression Significance
(1 – p-value)

Kendall’s Tau

Kendall’s Tau Significance
(percent)

R

R2

MEPDG standalone output

AB*

+0.24

98

0.42

95

0.71

0.50

ABAV*

+0.10

76

0.33

89

0.41

0.17

AD*

+0.03

54

0.33

89

0.27

0.07

AB, 100 mm**

-0.22

95

-0.60

88

-0.88

0.77

ABAV, 100 mm**

-0.41

96

-0.80

96

-0.90

0.80

AD, 100 mm**

-0.35

80

-0.60

88

-0.69

0.48

NF equation

AB*

0.19

44

0.33

89

0.21

0.05

AD*

0.25

61

0.33

89

0.30

0.09

AB, 100 mm**

-0.51

79

-0.40

76

-0.68

0.46

AD, 100 mm**

-0.38

79

-0.40

76

-0.68

0.46

1 mm = 0.039 inches

*n = 10
**n = 5
Note: AB = As-built, ABAV = As-built HMA with average unbound layers, and AD = As-designed.

Conclusions

This chapter considered a number of comparisons to answer the following two questions:

The change in the predicted rutting and fatigue cracking ranking among the different binders and thickness was assessed for as-built, as-built with average unbound layers, and as-designed scenarios. As-built inputs were faithful to the actual construction of the ALF lanes. As-built with average unbound layer inputs allowed only the impact of HMA construction to be assessed. As-designed inputs allowed the impacts of both HMA and unbound layer stiffness variability to be assessed. With regards to rutting, the ranking analysis indicates that the unbound layer properties did not have any significant influence on rutting. Analyses also indicated a very weak influence of HMA thickness and density (stiffness) on rutting. When this is considered in light of the statistical similarities of the measured rutting, the construction variability is of little to no concern. With respect to fatigue cracking, the same comparisons between the different scenarios indicate that HMA density (stiffness) has less influence over fatigue cracking than the unbound layer modulus. However, considering that the measured fatigue cracking performance had wide differences in crack initiation and propagation, construction variability is believed to only have a very weak influence on the ranked performance. The notable exception to this is the anomalous rutting performance of lane 6 terpolymer described in chapter 3.

Based on these analyses, when fatigue is compared to laboratory tests on mixture performance and binder parameters, the 4-inch (100-mm) sections are to be compared separately from the 5.8‑inch (150-mm) sections. This is because the 5.8-inch (150-mm) sections had fewer overall data points (five) and even fewer data points where sections exhibited cracking (three). Therefore, these sections can be used as qualitative check of rankings observed from the 4-inch (100-mm) cracking performance. When rutting is evaluated, all 4- and 5.8-inch (100- and 150‑mm) data points can be considered together since HMA thickness had less influence on rutting than cracking. However, the same division of data between the 4- and 5.8-inch (100 and 150-mm) ALF performance will be used to assess the laboratory mixture performance tests and binder parameters.

To summarize the evaluation of MEPDG predictive accuracy, the MEPDG seems to be able to tell differences between ALF performance when dissimilar conditions are combined together, such as different temperatures for rutting or different thickness for fatigue cracking. However, the MEPDG does not appear to differentiate between mixtures having only binder as the primary variable at a particular temperature or thickness. Some of this inaccuracy was expected because the ALF conditions were fairly extreme and polymer modified asphalts had very little representation in the national calibration. Finally, although both fatigue and rutting predictions were poor, the MEPDG seems to predict more practical fatigue cracking performance for the ALF conditions than the very impractical predicted rut depth.

 

Federal Highway Administration | 1200 New Jersey Avenue, SE | Washington, DC 20590 | 202-366-4000
Turner-Fairbank Highway Research Center | 6300 Georgetown Pike | McLean, VA | 22101