Download as pdf or txt
Download as pdf or txt
You are on page 1of 21

Modelling Flow Rutting in In-Service

Asphalt Pavements using the MechanisticEmpirical Pavement Design Guide


Erik Oscarsson
Department of Technology and Society
Lund University, Sweden
Box 118
221 00 Lund
Sweden
erik.oscarsson@tft.lth.se

ABSTRACT. Permanent deformation is a serious distress mechanism in asphalt pavements.


Modelling is a tool for performance prediction and deterioration assessment. This study
evaluated the Mechanistic-Empirical Pavement Design Guide (M-E PDG) model for
permanent deformation in asphalt concrete by employing the v1.003 software at level 1 and 3
on two Swedish motorway sections. Dynamic modulus data was obtained using the indirect
tensile test on field cores. Level 1 and 3 results differed heavily due to over-predicted dynamic
modulus data at level 3. However, the M-E PDG national field calibration is currently limited
to that level. Therefore, level 3 appears to be the only usable level until the predictive
dynamic modulus equation and the permanent deformation model have been further
calibrated. Previously found regional field calibration factors derived with accelerated
loading tests improved model accuracy, although national calibration was sufficient.
KEYWORDS:

Modelling, Permanent Deformation, Asphalt Pavement, M-E PDG.


DOI:10.3166/RMPD.12.37-56 2011 Lavoisier, Paris

Road Materials and Pavement Design. Volume 12 No. 1/2011, pages 37 to 56

38

Road Materials and Pavement Design. Volume 12 No. 1/2011

1. Introduction
Rutting due to permanent deformation is considered one of the most serious
distress mechanisms in asphalt pavements. It causes traffic hazards by affecting
vehicle steering. Further, an impervious road surface will trap water, snow and ice
that cause hydroplaning and loss of friction. Longitudinal cracks often occur in deep
ruts where they drain free water into the underlying pavement layers, thereby
increasing the deterioration rate. The factors affecting permanent deformations can
be divided into traffic loading, material properties and climatic conditions.
Modelling is a valuable tool used for pavement design and residue assessment. A
recently developed model is the incremental Mechanistic-Empirical Pavement
Design Guide (M-E PDG) (NCHRP, 2004), that considers the most important
distress types such as permanent deformation in bound, unbound and subgrade
layers, bottom-up and top-down fatigue cracking and transverse cracking. However,
this paper only considers permanent deformation in asphalt concrete (AC) layers,
which is also called flow rutting. The software simulation procedure involves
mechanistic calculation of stresses and elastic strains in each sublayer by using the
elastic multilayer subprogram JULEA, after which the plastic strains are assessed
with an empirical constitutive equation. The plastic strains in each increment are
then summed up into total permanent deformation (NCHRP, 2004). Some
advantages of the M-E PDG are its mechanistic-empirical framework, its
comprehensive calibration using LTPP data and its user-friendly software.
The M-E PDG concept is currently being implemented in the U.S. and could
prove useful in Sweden as well. Oscarsson (2007) verified the M-E PDG permanent
deformations model for asphalt concrete with data from an accelerated loading test
using Heavy Vehicle Simulator (HVS) applied on two representative Swedish
asphalt pavements. The model was considered applicable and regional field
calibration factors were derived.
The next step, i.e. the objective of this paper, was to validate the M-E PDG
permanent deformations model using in-service asphalt pavements, which is more
difficult. There are more contributing factors and larger variations to consider
compared to accelerated loading tests in which traffic and climate often are kept
constant for simplicity reasons. However, if reliable field data can be produced,
simulations using in-service roads have the advantage of being more realistic than
using accelerated loading tests.
This study is a qualitative field analysis based mainly on measured input data
rather than a quantitative regional calibration effort. Two asphalt pavement sections
on the Swedish motorway E6 were modelled using the M-E PDG software v1.003.
For asphalt materials, results from using input data accuracy level 1 and 3 were
compared, since they appear to produce different results (Azari el al., 2008), which
may be caused by over-prediction of the dynamic modulus at level 3 as shown by
previous studies (Dongr et al., 2005; Azari et al., 2007 and 2008; Zeghal et al.,

Modelling Flow Rutting using M-E PDG

39

2008). Further, previously found regional field calibration factors using accelerated
loading testing data on typical Swedish asphalt pavements derived by Oscarsson
(2007) was evaluated.

2. Methods and materials


The M-E PDG permanent deformation modelling was carried out using asphalt
materials data at both level 1 and 3 as they represent the highest and lowest input
data accuracy levels. Analysis using level 2 asphalt materials data was not carried
out as it is merely a mix of level 1 and 3 elements. All other data were constantly
input at the highest possible level with respect to the available field section
documentation that were collected by the Swedish National Road Administration
(SNRA, 1994) and the Swedish National Road & Transport Research Institute
Wiman et al., 1997). Thereby, the possible model result difference using level 1 and
3 asphalt materials input data was isolated. First, the model was evaluated using the
U.S. national calibration option, which means that regional calibration factors were
not used. Secondly, the previously found regional field calibration factors Er1=0.3,
Er2=1.5 and Er3=0.7 derived by Oscarsson (2007) using accelerated loading testing
data from typical Swedish asphalt pavements were employed and evaluated. Further,
level 1 measured and level 3 predicted dynamic modulus data were compared using
the M-E PDG software output data.

2.1. Structure and materials


Permanent deformation modelling was carried out on two test sections of the E6
motorway at Fastarp-Heberg, Sweden. The sections were selected due to their
significant amount of surface rutting. The relative field measurement error was
thereby minimized. The study focused on specific transversal sections rather than
mean values over distances. In practice, both field measurement and field coring for
specimen production were carried out in two specific transversal sections. Therefore,
the measured rut depth and material properties in a specific section are not
necessarily representative for all sections on the distance.
The semi-rigid pavement (CB) section 9:22, presented in Figure 1, was
constructed with a 42 mm Stone Mastic Asphalt (SMA) wear layer, 60 mm ABT
dense-graded asphalt concrete binder layer, 255 mm cement stabilized base layer, 65
mm unbound base layer, 640 mm crushed mineral aggregate sub base layer, 400 mm
coarse-grained subgrade layer and a semi-infinite layer of fine-grained subgrade.
The flexible pavement (GB) section 10:7 was constructed with a 35 mm SMA
wear layer, 187 mm AG dense-graded asphalt base layer, 88 mm unbound base
layer, 704 mm crushed mineral aggregate sub base layer, 400 mm coarse-grained
subgrade layer and a semi-infinite layer of fine-grained subgrade. Further details on
the E6 Fastarp-Heberg test sections are provided by Wiman et al. (1997).

40

Road Materials and Pavement Design. Volume 12 No. 1/2011

Figure 1. The layer structure of the test sections

2.1.1. Asphalt concrete layers


The properties of the asphalt concrete and its components according to SNRA
(1994) and Wiman et al. (1997) are presented in Table 1. Poissons ratio functions
for asphalt materials were assessed using the level 1 function within the M-E PDG
software limits.
The asphalt concrete dynamic compression modulus data used as M-E PDG
input were derived using the indirect tensile test (IDT). The standard uniaxial
dynamic modulus test using laboratory compacted specimens was not feasible as no
additional asphalt mix was available. Further, 150 mm thick field cores could not be
extracted from all layers due to limited layer thicknesses. Therefore, alternative
testing methods were considered. Kim et al. (2004) developed the IDT dynamic
modulus test and compared it with the uniaxial test. The comparison was carried out
using gyratory compacted specimens comprising 12 different asphalt mix types. The
data was used in a statistical analysis since they showed some variation. The unequal
variance t-test was performed for each mixture at five reduced frequencies. The
results showed that the data sets were statistically different in only 10% of the tests.
Therefore, IDT was considered applicable as a substitute for the uniaxial test. In this
study, the uniaxial test using gyratory compacted H150xD100 mm specimens was
replaced with the IDT test using H40xD150 mm field-cored specimens.

Modelling Flow Rutting using M-E PDG

41

Table 1. Properties of the asphalt concrete layers and its components


SMA
wear course

ABT asphalt
binder course

AG asphalt
base course

Stone mastic
asphalt

Dense-graded
hot mix asphalt

Dense-graded
asphalt base

16 mm

16 mm

22 mm

Cumulative retained mass on


19.05 mm (3/4 in) sieve

3%

3%

16%

Cumulative retained mass on


9.53 mm (3/8 in) sieve

58%

36%

42%

Cumulative mass retained on


4.75 mm (#4) sieve

74%

57%

57%

Mass passing 0.075 mm (#200)


sieve

8%

9%

5%

Bitumen softening point at 13000 P

47 C

47 C

38 C

Bitumen absolute viscosity at 60 C


(140 F)

1750 P

1750 P

601 P

Bitumen kinematic viscosity at


135 C (275 F)

377 cS

377 cS

202 cS

1.02 t/m3

1.02 t/m3

1.02 t/m3

Bitumen penetration at 25 C
(77 F)

85 (70/100)
dmm

85 (70/100)
dmm

190(160/220)
dmm

Bitumen Brookfield viscosity

N/A

N/A

N/A

Mix reference temperature

21.1 C

21.1 C

21.1 C

Mix effective binder content

6.2%

5.8%

4.2%

Mix air voids

4.8%

Mix name
Mix type
Max. aggregate size

Bitumen specific gravity at 25 C


(77 F)

Mix total unit weight


Mix thermal conductivity
Mix heat capacity

4.5%
3

5.2%
3

2.31 t/m

2.32 t/m

2.32 t/m3

0.35 mJ/h-m-C

0.35 mJ/h-mC

0.35 mJ/h-m-C

0.18 mJ/kg-C

0.18 mJ/kg-C

0.18 mJ/kg-C

Four 150 mm full-depth field cores were obtained from each of the semi-rigid
and flexible test sections. One specimen was cut from the upper portion of each
layer in each core. This resulted in four specimens per layer type. The stone mastic
asphalt (SMA) wearing layer of the flexible section cores was rejected due to an
average thickness of less than 35 mm. Instead, the material test results from the
theoretically identical semi-rigid section SMA specimens were used for both
sections. The two sections were subjected to the same traffic load and climate

42

Road Materials and Pavement Design. Volume 12 No. 1/2011

because they were located less than 1.5 km apart on a motorway with no junctions
between (Wiman et al., 1997). In addition, falling weight deflectometer data from
measurements carried out on the unbound base layer surfaces were similar. Finally,
both SMA layers were produced according to a single specification. However, the
sections differed regarding the asphalt binder and cement-treated base layers in the
semi-rigid section compared to the thick asphalt base layer in the flexible section.
Nonetheless, the differences were considered small enough to allow characterization
of both SMA layers using semi-rigid section SMA specimens only.
Three replicate specimens from each layer type were subjected to indirect tensile
haversine loading cycles at the temperatures -10 C, 5 C, 20 C and 35 C. At each
temperature, a frequency sweep was carried out using the frequencies 25, 20, 10, 5,
2, 1, 0.5, 0.2, 0.1 and 0.01 Hz. At each frequency, ten preconditioning pulses were
applied, followed by ten test pulses. In order to minimize the risk of excessive
deformation and invalid test results, no 0.01 Hz test at 35 C was carried out.
Further, a few invalid data points were removed because of their Poissons ratios
exceeding 0.5, which indicates specimen damage (Kim et al., 2004). The
temperature accuracy was 0.5 C measured with a thermometer integrated in a
dummy specimen. The strain level range was 40-60 microstrain in order to receive a
clear response signal while remaining within the linear viscoelastic range.
The dynamic modulus data was idealized using the sigmoidal function and
recalculated with respect to compaction method, aging and damage. The M-E PDG
requires data up to 55 C (131 F) although only data up to 35 C (95 F) were
available. Therefore, the sigmoidal function was used for extrapolation. This
procedure should not introduce any significant error due to the validity of the timetemperature superposition principle. Further, the pavement surface temperature at
the test site is above 35 C less than 0.02% of the time (Oscarsson and Said, 2010).
The M-E PDG is based on dynamic modulus data using gyratory compacted,
short-term aged specimens (NCHRP, 2004), while the test specimens were cored
from 12 years old in-service pavement layers. In order to consider compaction
differences, long-term aging and damage effects, the AG asphalt base dynamic
modulus data were compared with the corresponding IDT data derived using equally
composed specimens produced according to M-E PDG specifications by Oscarsson
(2007). At each temperature and frequency, the dynamic modulus ratio
|E*field specimens|/|E*laboratory specimens| was calculated. The compaction differences, longterm aging and damage effects appears to reduce dynamic modulus when low and
increase it when high, as can be seen in Figure 2. The ratios derived with AG asphalt
base specimens were then used to adjust the ABT binder and SMA wear layer
dynamic modulus data accordingly, as presented in Table 2. It should be noted that
the methodology assumes that the dynamic modulus of all layers are equally
affected by compaction differences, long-term aging and damage effects, regardless
of depth, air void content and binder properties. Figure 3 illustrates the range of test
data and the sigmoidal function for each mix type.

Modelling Flow Rutting using M-E PDG

IE*fieldl / IE*labl [MPa]

1,6
1,4
1,2
1
0,8
0,6
0

2000

4000

6000

8000

10000

12000

14000

16000

IE*labl [MPa]

Figure 2. Dynamic modulus ratio

Table 2. The adjusted dynamic modulus input data [MPa]


f [Hz]
Layer

SMA
wearing
layer

ABT
binder
layer

AG
ssphalt
base
layer

T [C]

0.1

0.5

10

25

-10

7523

9647

10459

12049

12607

13236

2205

3879

4759

6984

7946

11485

20

397

762

1024

2004

2620

3619

35

128

181

217

360

463

662

55

75

85

90

108

118

135

-10

7907

9617

10259

11513

11954

12455

1870

3236

3969

5884

6743

7853

20

376

677

886

1652

2133

2922

35

167

250

308

531

687

979

55

106

139

160

241

298

405

-10

8003

10079

10934

12752

13449

14282

2369

3727

4439

6327

7208

8399

20

599

1020

1283

2143

2637

3413

35

212

330

407

681

857

1161

55

98

129

148

215

259

336

43

44

Road Materials and Pavement Design. Volume 12 No. 1/2011

100000

lE*l [MPa]

10000

SMA measured data


SMA sigmoidal function
ABT measured data
ABT sigmoidal function
AG measured data
AG sigmoidal function

1000

100

10
-10

-8

-6

-4

-2

10

log fred [Hz]

Figure 3. Dynamic modulus test data and their sigmoidal functions


2.1.2. Cement stabilized base layer
The compressive yield stress of the cement stabilized base layer in the semi-rigid
section was stronger than planned (Wiman et al., 1997). The M-E PDG resilient
modulus input was derived at level 2 (NCHRP, 2004) using Equation [1]:

57000 f c'

[1]

The compressive strength fc=18.3 MPa reported by Wiman et al. (1997) was
geometrically adjusted from Kango cube (Andersson, 1987) results to AASHTO
(2007) T22-07 standard results using Swedish standards (2005) S 137207
conversion factors. Equation [1] resulted in a reasonable resilient modulus of 17400
MPa (2530 ksi), which is 26% higher than the default value. Default minimum
resilient modulus and modulus of rupture were also adjusted accordingly, i.e. 869
MPa (126 ksi) and 5661 kPa (821 psi), respectively.
2.1.3. Unbound layers
For both pavements, the unbound base layer, the crushed mineral aggregate sub
base layer and most of the coarse-grained subgrade layer were joined in a single
1054 mm (41.5 in) A-1-a layer according to the AASHTO (2007) M145-91 soil
classification. As the thick pavement sections reached the softwares maximum
allowed number of sublayers (NCHRP, 2004), a part of the coarse-grained materials,
i.e. 51 mm (2.0 in) for the semi-rigid section and 138 mm (5.4 in) for the flexible
section, had to be merged with the semi-infinite A-5 layer characterizing the finegrained subgrade. The necessary simplification should not significantly affect
results, as doubling the error only resulted in 0.003 mm and 0.03 mm extra rut depth
in 10 years for the semi-rigid and the flexible section, respectively. The unbound
materials were characterized using the M-E PDG representative default values at
level 3 due to a lack of measured data. Consequently, a Poissons ratio of 0.35 and a
coefficient of lateral pressure of 0.5 were used for both pavements. The default

Modelling Flow Rutting using M-E PDG

45

software resilient modulus was 248 MPa (36 ksi) and 83 MPa (12 ksi) for the A-1-a
and A-5 materials, respectively (NCHRP, 2004).

2.2. Traffic
Traffic properties were input at level 1. The number of vehicles was assessed
using pneumatic tubes or inductive loops mounted on the road surface, while the
heavy axle load spectrum was characterized using Weigh-In-Motion (WIM) data
from SNRA (2008a) as described by Winnerholt (2004). The available one-week
early summer WIM measurements were carried out in 2004 near Kungsbacka and
2005-2007 near Lddekpinge. These locations are 95 km north and 130 km south
from test sections, respectively. The heavy vehicle load spectra in Kungsbacka and
Lddekpinge were considered similar enough to allow interpolation of data.
The number of heavy vehicles was directly measured in the southbound right
lane using pneumatic tubes or inductive loops by SNRA (2008b). However, as the
equipment erroneously included light vehicles with long axle spacing, the results
were adjusted using the more detailed WIM data. The resulting initial two-way
heavy vehicle average annual daily truck traffic (AADTT) of 1559 heavy vehicles
per day and the mean compound growth was 5.20%. The mean operational speed
calculated using WIM data was found to be constant at 88.7 km/h (55.1 mph), as
presented in Table 3.

Table 3. General traffic input data


Initial two-way heavy vehicle traffic
No. of lanes in design direction

1559 vehicles/day
2

Percent trucks in design direction

50%

Percent trucks in design lane

100%

Operational speed

88.7 km/h (55.1 mph)

Compound growth

5.20%

M-E PDG uses the assumption of constant axle load spectrum from year to year,
over the week and throughout the day (NCHRP, 2004). In this study, the axle load
distributions were also assumed constant throughout each year since each WIM
measurement covered only one week. The assumption appears reasonable as
Lu et al. (2002) reported very small seasonal load spectra variation in California
despite the addition of agricultural transports during harvest time. Therefore, default
monthly adjustment factors were used. The distribution of heavy vehicles
throughout the day was assessed using the WIM data presented in Figure 4.

Road Materials and Pavement Design. Volume 12 No. 1/2011

Hourly distribution

46

7%
6%
5%
4%
3%
2%
1%
0%
0

12

18

24

Hour of the day


Figure 4. Distribution of heavy vehicles throughout the day

The heavy vehicles of the interpolated WIM data were classified according to the
FWHA vehicle classification (NCHRP, 2004). Separation of certain infrequent
vehicle types was difficult and offered limited benefits. Therefore, class 4 vehicles
were absorbed by classes 5 and 6 while class 10 vehicles joined class 12. The
distribution of heavy vehicles into vehicle classes is shown in Table 4, where also
the average number of axles per truck is defined.

Table 4. The class distribution of heavy vehicles and number of axles per truck
Number of axles per truck

Vehicle class

Distribution of
heavy vehicles

Single

Tandem

Tridem

0.00%

19.1%

11.3%

1.7

0.65

0.5%

0.76

0.64

0.66

6.3%

27.0%

1.96

0.09

0.96

10

0.00%

1.33

3.34

11

10.1%

3.56

0.45

12

11.1%

2.12

1.24

0.47

13

14.6%

2.23

2.09

0.25

Axle load distribution factors for each vehicle class based on WIM data are
presented in Figure 5. The typical single axle range was 25-100 kN, 50-200 kN for
tandem axles and 50-250 kN for tridem axles. Quad axles were not used since the
few WIM recorded quad axles were approximated as tridem axles.

Modelling Flow Rutting using M-E PDG

47

100
90

Accumulated distribution

80

1A, VC 5
1A, VC 6
1A, VC 7
1A, VC 8
1A, VC 9
1A, VC 11
1A, VC 12
1A, VC 13
2A, VC 6
2A, VC 7
2A, VC 8
2A, VC 9
2A, VC 11
2A, VC 12
2A, VC 13
3A, VC 7
3A, VC 9
3A, VC 12
3A, VC 13

70
60
50
40
30
20
10
0
0

50

100

150

200

250

300

350

Axle load [kN]

Figure 5. Axle load distribution factors for single axles (1A), tandem axles (2A) and
tridem axles (3A) in each vehicle class (VC)

The used axle configuration data presented in Table 5 were default values
(NCHRP, 2004). Assessment of the mean wheel location was based on the design
lane width and the average axle width. The lateral wander average standard
deviation was estimated using the lane width and results by Buiter et al. (1989).

Table 5. Axle configuration and lateral traffic wander input data


Average axle width (edge-to-edge outside dimensions)

2.59 m (8.5 ft)

Dual tire spacing

0.30 m (12 in)

Tire pressure

828 kPa (120 psi)

Tandem axle spacing

1.31 m (51.6 in)

Tridem axle spacing

1.25 m (49.2 in)

Quad axle spacing

1.25 m (49.2 in)

Mean wheel location from the lane marking [in]

0.46 m (18 in)

Traffic wander standard deviation [in]

0.29 m (11.4 in)

Design lane width [ft]

3.50 m (11.5 ft)

48

Road Materials and Pavement Design. Volume 12 No. 1/2011

2.3. Climate
An integrated climatic model (icm) file at input accuracy level 1 was created
using hourly climatic data (hcd) from the Swedish Meteorological and Hydrological
Institute (SMHI, 2008) and the Swedish Road Administration (SNRA, 2008c). The
climatic data time interval comprised January 2003 to February 2006,
i.e. 38 months, as compared with the M-E PDG minimum requirement of 24 months
(NCHRP, 2004).
The ground water depth, calculated as the mean value of the nearby Halmstad
and Oskarstrm measurements (SGU, 2008), was constant throughout the year. The
daily solar radiation maximum and sunrise/sunset times were modelled as a function
of latitude (NCHRP, 2004). Air temperature, wind speed and relative humidity were
provided by hourly on-site measurements (SNRA, 2008c). Precipitation was
approximated by accumulated 6 h data from Halmstad distributed hourly according
to Helsingborg 1 h data. The sunshine percent was approximated using Vxj data
as that was the only nearby and complete sunshine data set (SMHI, 2008). A
summary of the climate input data is provided in Table 6.

Table 6. Climate input data summary


Longitude, latitude

12.46E, 56.47N

Elevation

18 m (60 ft)

Annual water table depth

2.2 m (7.2 ft)

Mean annual temperature

7.2 C (45 F)

Number of freezing degree days


Annual rainfall
Average relative humidity
Average wind speed

420
806 mm (31.7 in)
81%
3.5 m/s (7.7 mph)

Solar radiation minimum

825 W/m2 (261 BTU/ft2-h)

Solar radiation maximum

10700 W/m2 (3380 BTU/ft2-h)

2.4. Field measurement


The transversal surface profile was measured basically twice per year in April
and October from 1997 to 2006 using a PRIMAL profilometer (Wiman et al., 2009).
Figure 6 shows the total rut depth data derived from the measured profile data using
the wire line method. M-E PDG rut depth results can be compared directly to these
measurements as they were calibrated using LTPP data derived with the same

Modelling Flow Rutting using M-E PDG

49

method (NCHRP, 2004). However, during winter Swedish asphalt pavements


exhibit considerable wear rutting due to studded tires as reported by Wiman et al.
(2009), while M-E PDG show virtually no rutting increment. Therefore, the total rut
depth data was adjusted for wear rutting using additional laser profilometer
measurement data (Wiman et al., 2009), which resulted in the rut depths presented
in Figure 7. Again, it should be noted that the specific sections selected for this
study are not necessarily representative for the layer structures and materials.
Comparisons between observation and model results were not carried out after a
specific time. Instead, the comparison was carried out using mean values of all
observation points in order to account for measurement variation.
Date
2000-01-01

Date
2004-01-01

2008-01-01

1996-01-01
0

Rut depth [mm]

Rut depth [mm]

1996-01-01
0

6
8
10
12
14
16
18
20

Semi-rigid section
Flexible section

Figure 6. Total observed rut depth in


the semi-rigid and flexible test
sections

2000-01-01

2004-01-01

2008-01-01

6
8
10
12
14
16
18
20

Semi-rigid section
Flexible section

Figure 7. Observed rut depth in the


semi-rigid and flexible test sections
adjusted for wear rutting

3. Results
3.1. Simulation using the national calibration option
Simulations using the national calibration option at level 1, presented in
Figures 8 and 9, showed a 135% and 38% mean over-prediction in total rut depth for
the semi-rigid and flexible sections, respectively. At level 3, the model overpredicted the semi-rigid section by 18%, while the flexible section was underpredicted by 22% as illustrated in Figures 10 and 11. Level 3 results appeared to
correspond better to observation than level 1 results.
It was also noted that the modelled contribution from unbound base and
subgrade layers compared to the total observed deformation was very high even
after ten years of service. At level 1 it was 43% and 57%, and at level 3 it was 39%
and 46% for the semi-rigid and flexible sections, respectively. In Sweden, that
should be regarded as high considering the thick layers of asphalt concrete covering
the high quality crushed rock material layers (Wiman et al., 2009).

50

Road Materials and Pavement Design. Volume 12 No. 1/2011

Figure 8. Rut depth in the semi-rigid


test section using national calibration
at level 1

Figure 9. Rut depth in the flexible test


section using national calibration at
level 1

Figure 10. Rut depth in the semi-rigid


test section using national calibration
at level 3

Figure 11. Rut depth in the flexible


test section using national calibration
at level 3

3.2. Simulation using previously found calibration factors


Permanent deformation modelling at level 1 for the field sections using
calibration factors derived by accelerated testing suffered from heavy overpredictions by 245% and 82% for the semi-rigid and flexible sections as presented in
Figures 12 and 13. At level 3, the model over-predicted the semi-rigid section by
47%, while the flexible section was under-predicted by 9% as illustrated in
Figures 14 and 15. Again, level 3 results appeared to correspond better to
observation than level 1 results. The previously noted considerable contribution of
modelled permanent deformations from unbound base and subgrade layers to
observed total rut depth was, as expected, not affected by use of regional field
calibration factors for asphalt concrete layers.

Modelling Flow Rutting using M-E PDG

51

Figure 12. Rut depth in the semi-rigid Figure 13. Rut depth in the flexible test
test section using calibration constants section using calibration constants from
from HVS test at level 1
HVS test at level 1

Figure 14. Rut depth in the semi-rigid Figure 15. Rut depth in flexible test
test section using calibration constants section using calibration constants from
from HVS test at level 3
HVS test at level 3

3.3. Comparison of predicted and measured dynamic modulus data


Comparison of level 1 measured and level 3 predicted dynamic modulus data
showed heavy dynamic modulus over estimations by the predictive equation, as
illustrated in Figures 16 and 17. Dynamic modulus data, derived with the M-E PDG
software, is presented monthly during a normally tempered year, i.e. at the 3rd
temperature quintile. All dynamic modulus values are relatively high due to the high
loading frequency at motorway speed. The ratio of |E*level 3|/|E*level 1| in the semirigid section ranged from 1.95 to 3.05 with a mean value of 2.36, while the flexible
section ratio ranged from 1.95 to 3.52 with a mean value of 2.61. The results were
similar in all sublayers that were located at different depths in the pavement.

52

Road Materials and Pavement Design. Volume 12 No. 1/2011

Figure 16. Comparison of M-E PDG


predicted and measured dynamic
modulus data in AC sublayers at
different depths in the semi rigid
pavement section

Figure 17. Comparison of M-E PDG


predicted and measured dynamic
modulus data in AC sublayers at
different depths in the flexible avement
section

4. Discussion
The difference in permanent deformation results using material input data at
level 1 and 3 can only be explained by material properties, as traffic and climate
were input at level 1 in both cases. The main difference between level 1 and 3
material input data is the use of the M-E PDG predictive dynamic modulus equation,
which is sometimes referred to as the Witczak predictive equation. In this study, the
predicted dynamic modulus was generally found to be more than twice the measured
dynamic modulus. However, the results of this paper relied on a number of
assumptions. First, all dynamic modulus data of the long-term aged, field compacted
and possibly damaged specimens were recalculated according to the dynamic
modulus derived by Oscarsson (2007) using IDT with gyratory compacted AG
asphalt base specimens. Second, the dynamic modulus data derived with the IDT
was assumed similar to that of the uniaxial test, as statistically shown by Kim et al.
(2004). Thirdly, the number gap-graded mixes included in the M-E PDG calibration
was very limited (NCHRP, 2004). Therefore, NCHRP (2004) recommends that the
SMA materials should primarily be characterized at input level 1 only. The
relevance of comparison between level 1 and 3 dynamic modulus is thereby limited
for SMA layers, i.e. the three first sublayers in Figure 16 and 17. However, both the
ABT and AG materials show the same behaviour.
Previous studies have reported that the predictive dynamic modulus equation
tends to over-predict at low values, i.e. at high temperature and/or low loading

Modelling Flow Rutting using M-E PDG

53

frequency. Dongr et al. (2005) found that although the model results were
reasonable at low temperatures, it should not be used at high temperatures, i.e. when
the dynamic modulus is below 689 MPa (100 ksi). Using a variety of binders and
compaction methods, Azari et al. (2007) also concluded that high modulus
predictions are reasonable but that at low modulus values the results tend to be overpredicted as illustrated in Figure 18, which can be compared with the results of this
study in Figures 16 and 17. Based on a wide selection of mix types, Zeghal and
ElHussein (2008) found that the predictive model under-predicted at low
temperatures and over-predicted at high temperatures, especially at low loading
frequency. In order to minimize the difference between level 1 and 3 permanent
deformation results, this study recommends that the dynamic modulus prediction
model should be further refined, especially at high temperature and low loading
frequency.

Figure 18. Comparison of measured and NCHRP 1-37Apredicted |E*| values for
various binders (Azari et al., 2007)

The M-E PDG permanent deformations model results appeared to produce better
predictions at level 3 than at level 1 when the opposite should theoretically be true.
However, as Azari et al. (2008) suggested, the model is indeed nationally calibrated
using material input data at level 3 according to the M-E PDG documentation in
Appendix EE (NCHRP, 2004). As previously discussed, the predictive dynamic
modulus equation used for level 3 calibration has been found to over-predict at high
temperature and low loading frequency, which are the conditions under which
permanent deformations typically occur. Therefore, there is a considerable risk that
the level 3 permanent deformation model calibration is based on over-predicted
dynamic modulus data. If so, the nationally calibrated model must be over-sensitive

54

Road Materials and Pavement Design. Volume 12 No. 1/2011

in order to produce results corresponding to field observations in the long-term


pavement performance (LTPP) database. That would result in over-prediction when
using measured dynamic modulus input data at level 1, which explains the results of
Azari et al. (2008) and this study. Further, it appears logical to use the model only
up to the accuracy level it is calibrated for. Therefore, this study recommends use of
material input data at level 3 only, until further calibration efforts have been carried
out.
The regional calibration factors Er1=0.3, Er2=1.5 and Er3=0.7 derived by
Oscarsson (2007) somewhat increased the modelled permanent deformations.
Level 1 analysis generally resulted in over-predictions, while level 3 corresponded
fairly well with observations. The modelled contributions from unbound base and
subgrade to total observed permanent deformation were larger than expected in the
robust test sections. The contribution of the M-E PDG permanent deformation
models to total rut depth is calibrated using a trench study comprising only seven
MnRoad pavement sections. The well-known need for further calibration based on
trench data (NCHRP, 2004; NCHRP, 2006; Oscarsson, 2007) is therefore
emphasized.

5. Conclusions
Based on material testing and modelling permanent deformations in a semi-rigid
and a flexible in service pavement section using the M-E PDG software v1.003, the
following conclusions were drawn:
According to this study, material data for permanent deformation modelling
should only be input at level 3 until calibration at higher levels has been carried out.
Before performing further calibration, the predictive dynamic modulus
equation used at level 2 and 3 should be refined in order to reduce model result
differences between the three input levels.
Use of the regional field calibration factors Er1=0.3, Er2=1.5 and Er3=0.7,
previously found by accelerated testing, yielded reasonable results, as did national
calibration.
The modelled distribution of permanent deformation between layers should be
further calibrated using additional trench studies.
Acknowledgements
The financial support provided by the Swedish National Road Administration
and the Development Fund of the Swedish Construction Industry is gratefully
acknowledged. Further, the project relied on supervision and technical support by
Dr. Safwat Said at the Swedish National Road and Transport Research Institute,
Dr. Monica Berntman at Lund University and Dr. Richard Nilsson at Skanska
Teknik.

Modelling Flow Rutting using M-E PDG

55

6. Bibliography
AASHTO, AASHTO Provisional Standards, 2007 Edition, Washington D.C., American
Association of State Highway and Transportation Officials, 2007.
Andersson R., Belggningar av vltbetong: Provningsmetoder och kontroll, In Swedish,
Report 2-87, 1987, Swedish Cement and Concrete Research Institute.
Azari H., Al-Khateeb G., Shenoy A., Gibson N., Comparison of Simple Performance Test
E* of Accelerated Loading Facility Mixtures and Prediction E*, Transportation
Research Record, Vol. 1998, 2007, p. 1-9.
Azari H., Mohseni A., Gibson N., Verification of Rutting Predictions from Mechanistic
Empirical Pavement Design Guide using Accelerated Loading Facility Data,
Transportation Research Record, Vol. 2057, 2008, p. 157-167.
Buiter R., Cortenraad W.M.H., van Eck A.C., Van Rij H., Effects of transverse distribution
of heavy vehicles on thickness design of full-depth asphalt pavements, Transportation
Research Record, Vol. 1227, 1989, p. 66-74.
Dongr R., Myers L., DAngelo J., Paugh C., Gudimettla J., Field Evaluation of Witczak and
Hirsch Models for Predicting Dynamic Modulus of Hot-Mix Asphalt, Journal of the
Association of Asphalt Paving Technologists, Vol. 74, 2005, p. 381-442.
Kim Y.R., Seo Y., King M., Momen M., Dynamic modulus testing of asphalt concrete in
indirect tension mode, Transportation Research Record, Vol. 1891, 2004, p. 163-173.
Lu Q., Harvey J., Le T., Lea J., Quinley R., Redo D., Avis J., Truck Traffic Analysis using
Weigh-In-Motion (WIM) Data in California, Draft report, June 2002, Pavement
Research Center, University of California at Berkeley.
NCHRP, Guide for the Mechanistic-Empirical Design of New and Rehabilitated Pavement
Structures, Report 1-37A, National Cooperative Highway Research Program, 2004.
NCHRP, Independent Review of the Mechanistic Empirical Pavement Design Guide and
Software, Research Results Digest 307, National Cooperative Highway Research
Program, 2006.
Oscarsson E., Prediction of Permanent Deformations in Asphalt Concrete using the
Mechanistic-Empirical Pavement Design Guide, Licentiates Thesis, Bulletin 236,
Department of Technology and Society, Lund University, Lund, 2007.
Oscarsson E. and Said S., Modeling permanent deformations in asphalt concrete layers using
the PEDRO model, Working paper, 2010.
SGU, Ground water depth data from Halmstad and Oskarstrm 1998-2008, The Geological
Survey of Sweden, 2008.
SMHI, Licensed Swedish meteorological data 1996-2008, Swedish Meteorological and
Hydrological Institute, 2008.

56

Road Materials and Pavement Design. Volume 12 No. 1/2011

SNRA, Vg 94: General technical specifications for road structures, Publ. 1994:26, In
Swedish, 1994, Swedish National Road Administration.
SNRA, WIM data: Kungsbacka 13-536 southbound 2004 and Lddekpinge M482
southbound 2005-2007, Swedish National Road Administration, 2008a.
SNRA, Traffic flow data sets 454 and 9454 from Klickbara Kartan web data base, Swedish
National Road Administration, 2008b. Accessed 2008-05-28.
SNRA, Data from VViS station No. 1336, Swedish National Road Administration, 2008c.
Swedish standards (SS-EN), Swedish Standards Institute, 2005.
Wiman L.G., Hermelin K., Ydrevik K., Hultqvist B-., Carlsson B., Viman L., Eriksson L.,
Prov med olika verbyggnadstyper: Observationsstrckor p vg E6, Fastarp-Heberg,
In Swedish, Report 56:1, The Swedish National Road & Transport Research Institute,
1997.
Wiman L.G., Carlsson H, Viman L, Hultqvist B-., Prov med olika verbyggnadstyper.
Uppfljning av observationsstrckor p vg E6 Fastarp-Heberg, 10-rsrapport, 19962006, In Swedish, Report 632, The Swedish National Road & Transport Research
Institute, 2009.
Winnerholt T., BWIM-mtningar 2002 och 2003, In Swedish, Report 2003:165, Swedish
National Road Administration, 2004.
Zeghal M., ElHussein H.M., Assessment of analytical tools used to estimate the stiffness of
asphalt concrete, Canadian Journal of Civil Engineering, Vol. 35, No. 3, 2008, p. 268275.

Received: 25 May 2009


Accepted: 30 July 2010

Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.

You might also like