Methodology to Predict Alligator Fatigue

Cracking Distress Based on Asphalt

Concrete Dynamic Modulus
Sherif M. El-Badawy, Myung Goo Jeong, and Mohamed El-Basyouny

This study focused on integrating the MechanisticEmpirical Pavement obvious that, if these parameters could be linked to an accurate pave-
Design Guide (MEPDG) methodology (NCHRP 1-37A and NCHRP ment performance prediction model, a truly rational and fundamentally
1-40D) with the simple performance test methodology (NCHRP 9-19) to based pay factor system could be developed.
develop a comprehensive bottom-up fatigue cracking distress prediction NCHRP Projects 1-37A and 1-40D produced the Mechanistic
model based on the dynamic modulus in a methodology that can be Empirical Pavement Design Guide (MEPDG) to predict pavement
implemented in a probabilistic performance-related specification (PRS) performance (26). MEPDG is a very powerful tool for the analysis
methodology for quality assurance (QA) of hot-mix asphalt (HMA) con- and design of new and rehabilitated pavements. However, it requires
struction (NCHRP 9-22). The main approach for a comprehensive fatigue an excessive amount of computation time to complete stochastic
crackingdamage predictive methodology was to run MEPDG for a large simulations if implemented because of the principle of stochastic
number of simulations using combinations of inputs that were believed analysis. As an alternative tool to save the time and simultaneously
significant in classic load-associated bottom-up alligator fatigue cracking to remain relatively close to the MEPDG solution in terms of imple-
and to develop an accurate closed-form solution for the fatigue damage mentation of performance-related specifications (PRS), develop-
cracking distress prediction based on the dynamic modulus of HMA. ment of a comprehensive fatigue cracking prediction methodology
Four major studies have been completed for fatigue cracking prediction was critical.
methodology: (a) development of a fatigue damage model based on a MEPDG Version 1.0 was run for a large number of simulations
two-layer pavement system, (b) development of an asphalt concrete (AC) using various combinations of inputs that are considered the most
effective dynamic modulus model, (c) development of a predictive method- signicant variables in classic bottom-up alligator fatigue cracking
ology to transform a multilayer to a two-layer pavement system, and to develop an accurate closed-form solution for the fatigue damage
(d) validation of the prediction process. The developed methodology cracking distress prediction. Four major studies have been com-
approximately but accurately predicts the classic AC alligator fatigue pleted for development of the fatigue cracking prediction method-
cracking distress that is close to that from MEPDG. Thus, implementation ology: (a) development of a bottom-up fatigue damagecracking
of the methodology into the PRS-based QA system in NCHRP 9-22 is model, (b) development of an AC effective dynamic modulus model,
recommended. (c) development of a predictive methodology to estimate the com-
posite foundation modulus from the actual layered pavement system
below the AC layer, and (d) validation of the overall prediction
In current department of transportation construction practice, asphalt process.
concrete (AC) acceptance, as a constructed-in-place material, is The developed methodology can then be used in a probabilistic
based on quality assurancequality control parameters such as asphalt approach that utilizes the mean and standard deviation for the vari-
content, air voids, and job mix formula gradation. For all practical ables in the fatigue cracking prediction. Thus, the mean and standard
purposes, each of these quality assurancequality control parameters deviation of the alligator fatigue cracking distribution can be used as
has a set of pay factors that have been empirically established over the the basis for a stochastic PRS analysis for the quality assurance (QA)
years through the judgment of the pavement community. The pay of hot-mix asphalt (HMA) construction.
factor has been used to reward or to penalize contractors for material
and construction quality on a given project (1).
No known accurate linkage exists between pay factor magnitude OBJECTIVES
and performance or actual anticipated pavement life. Therefore, it is
The primary objective of this research is to integrate MEPDG
(NCHRP 1-37A and NCHRP 1-40D) and simple performance test
(SPT) (NCHRP 9-19) methodologies to develop a bottom-up fatigue
S. M. El-Badawy, Public Works Engineering Department, Faculty of Engineering,
Mansoura University, Mansoura, Egypt. M. G. Jeong and M. El-Basyouny, Department cracking distress prediction model based on the dynamic modulus
of Civil and Environmental Engineering, Arizona State University, P.O. Box 875306, (E*) in a methodology that can be implemented in a stochastic PRS
Tempe, AZ 85287-5306. Corresponding author: M. G. Jeong, [email protected]. methodology for QA of HMA construction (NCHRP 9-22). The major
advantage of the proposed methodology is that a probabilistic solution
Transportation Research Record: Journal of the Transportation Research Board,
No. 2095, Transportation Research Board of the National Academies, Washington,
can be determined in a matter of minutes for a given construction
D.C., 2009, pp. 115124. problem compared with multiday solutions when MEPDG is used
DOI: 10.3141/2095-12 as the primary software code source.

115
116 Transportation Research Record 2095

BOTTOM-UP FATIGUE Design Traffic

DAMAGECRACKING METHODOLOGY
All the simulation runs were done by using the classic 18-kip
Pavement Structure equivalent single-axle load (ESAL) approach. The number of
traffic repetitions used in the analyses was set at 2 106 ESALs
The pavement structure selected for the simulation runs consists of in 20 years.
one HMA layer over a foundation layer with a composite modulus.
This composite foundation is a representative subsurface layer having
one modulus that represents different types of subsurface layers that Vehicle Speed
may be encountered in practice.
Four traffic speeds, ranging from creep speed to highway speed,
were used in the study. These speeds are as follows:
Design Parameters
0.5 mph: intersections, parking lots, traffic jams;
Seven factors were considered for the MEPDG simulations. The 15 mph: local roads, school zones;
selected factors were believed to affect fatigue-related distress sig- 45 mph: collector and arterial city roads; and
nicantly. Table 1 shows the full factorial of the simulations. It also
60 mph: city freeways and Interstate highways.
summarizes the levels used for each factor. The following subsections
explain the levels of each variable.

AC Layer Thickness
Environmental Sites Seven AC layer thicknesses were used to cover a wide range of AC
thicknesses used in practice. They were selected to ensure that the
Three climatic locations were selected for the simulations. They were
chosen to cover a broad range of temperature conditions as follows: extreme fatigue conditions (controlled strain and controlled stress)
would be assessed in the study.
Cold: (Grand Forks, North Dakota), mean annual air temperature
(MAAT) = 40.4F;
Moderate: (Oklahoma City, Oklahoma), MAAT = 60.5F; and Asphalt Mix Characteristics
Hot: (Key West, Florida), MAAT = 77.8F.
Three levels of asphalt binder performance grades were selected
according to the Superpave system: PG 82-10, PG 64-22, and
Design Life PG 52-40. These grades provide cover ranging from the very soft
binder (PG 52-40) typically used in a region with a cold climate
A design life of 20 years was selected for the simulation. through the very stiff one (PG 82-10) used in a hot region. Addition-

TABLE 1 Matrix of Fatigue Distress MEPDG Simulation Runs

Number
Factor Value of Values

Environmental sites Grand Forks, North Dakota 3

Oklahoma City, Oklahoma
Key West, Florida
Design life (years) 20 1
Design traffic (ESALs) 2  106 1
AC thickness (in.) 1, 2, 3, 4, 6, 8, and 20 7
Vehicle speed (mph) 0.5, 15, 45, and 60 4
E* of AC Aggregate gradation [34-in. nominal 1
aggregate size (retained 34 in. = 0%,
retained 38 in. = 26.8%, retained
No. 4 = 46.41%, passing No. 200 = 8.28%)]
PG grades (PG 82-10, PG 64-22, 3
and PG 52-40)
Va (air voids) and Vb (binder content): 3
(Va = 4, Vb = 16), (Va = 8, Vb = 12),
and (Va = 12, Vb = 8)
Ecf (composite foundation 3, 8, 25, 75, 250, and 1,000 6
modulus) (ksi)

NOTE: Total number of runs performed = 4,536 runs (1,512/site).

El-Badawy, Jeong, and El-Basyouny 117

ally, three levels of voids filled with binder (VFB) and one set of Vb
M = 4.84 0.69 (5)
gradation were used. The combination of these mix characteristic Va + Vb
levels contributed to changing E* which ranged from approximately
30 to 3,000 ksi for the different AC layer thicknesses and vehicle where
speeds used for the simulations.
t = tensile strain at the critical location;
e = stiffness of material, psi;
Composite Foundation Modulus f1, f 2, f3 = national calibration parameters (1.0, 1.2, and 1.5,
respectively);
Fatigue cracking is greatly affected by foundation stiffness. Thus, hac = thickness of asphalt layer, in.;
the individual thicknesses and moduli of all layers beneath the HMA Vb = effective binder content, %; and
layers inuence the fatigue behavior of the AC mixture; the imaginary Va = air voids, %.
foundation stiffness combining all the bottom layers moduli is called
In the calibrated version of MEPDG; fatigue cracking is trans-
composite foundation modulus (Ecf ). This fact necessitated running
formed into damage, knowing the AC layer thickness by using
a matrix of different Ecf values to cover a broad range of unbound
Equation 6 (7, 8):
bound, basesubbase, and subgrade materials that may represent
the foundation for the pavement structure in the eld. Six Ecf values
6, 000 1
covering soft to hard composite foundations were used for the study. FC =
1 + e(C1 C1+C2 C2 log 10( D )) 60
(6)
A subsystem methodology was then developed to take any unique
combination of basesubbasesubgrade thickness and stiffness
(modulus) values and then predict what the composite foundation where
moduli would be. The details of the developed methodology are
presented later. C1 = C2 = 1.0,
C1 = 2 * C2,
C 2 = 2.40874 39.748 (1 + hac)2.856, and
Database Analysis and Results FC = fatigue cracking (% of lane area).

This section discusses the set of analyses and results of the 4,536 com-
puter simulation runs performed for this study using MEPDG Ver- Relationship Between Design Traffic (ESALs)
sion 1.0. All the simulation runs were performed based on the design and Damage
shown in Table 1. In addition, the analyses were conducted using the
percentage alligator (bottom-up) fatigue damage instead of cracking. A ministudy was initially performed before the Table 1 matrix
Cracking reects eld performance data used to correlate to damage analysis to validate the hypothesis that damage and traffic repeti-
that theoretically can be computed as a function of strain, stress, tions were directly correlated. A total of 34 MEPDG simulation runs
deflection, and so forth (5). Damage is defined as the ratio of the were conducted with different AC thicknesses, traffic speeds, pave-
predicted number of traffic repetitions to the allowable number of ment design life values, Superpave binder grades, effective binder
load repetitions (to some failure level) as shown in Equation 1. contents, air voids, and composite foundation moduli. There was a
direct relationship between the number of traffic repetitions and
T damage, if all other variables are held constant. This relationship can
D=
ni
(1) be expressed as follows:
i =1 N fi

T
where Di = D j ri (7)
Trj
D = fatigue damage,
T = total number of computational periods,
where Di is percent damage predicted at traffic repetitions Tri, and Dj
ni = actual traffic for period i, and
is percent damage predicted at traffic repetitions Trj.
Nfi = allowable failure repetitions under conditions prevailing in
period i.
The allowable number of repetitions to failure (Nf ) is calculated in Environmental Location
MEPDG methodology by using the following set of equations (7, 8):
The inuence of the environmental location was found to be one of the
3.291 f 2 0.854 f 3 most signicant factors affecting development of the alligator fatigue
1 1
N f = 0.007566 C f 1 k1
E
(2) cracking model. To implement the environmental location into a
t
regression model to predict alligator fatigue damage, each location
was expressed by a climatic factor. Details of the development of the
1
k1 = (3) climatic factor equation can be found elsewhere (4, 911).
0.003602
0.000398 +
1 + e(11.02 3.49 hac )
T f = 1.0056 ( MAAT ) + 0.8755 ( MMAT ) 1.1861( wind )
C = 10 + 0.5489 ( sun ) + 0.0706 ( rain )
M
(4) (8)
118 Transportation Research Record 2095

where 1.0E+03
T f = climatic factor, F; 1.0E+02
MAAT = mean annual air temperature, F; 1.0E+01

Fatigue Damage (%)

MMAT = standard deviation of mean monthly air temperature in 1.0E+00
a given year, F; 1.0E-01
wind = mean annual wind speed, mph; 1.0E-02
sun = mean annual sunshine, %; and 1.0E-03
rain = mean cumulative rainfall depth, in. 1.0E-04
1.0E-05
1.0E-06
Composite Foundation Modulus
1.0E-07
Analysis of the simulated data showed that Ecf was a very signicant 0 5 10 15 20
variable affecting fatigue performance. As this value increased, AC Layer Thickness (in.)
alligator fatigue damage decreased. Furthermore, the relationship
Ecf = 3 ksi Ecf = 8 ksi Ecf = 25 ksi
between fatigue damage and Ecf was found to follow a second-degree
polynomial function in log space. Ecf = 75 ksi Ecf = 250 ksi Ecf = 1000 ksi

FIGURE 1 Influence of AC thickness on damage (VFB 0.8,

log D = a1 log ( Ecf ) + a2 log ( Ecf ) + a3
2
(9) PG 64-22, 0.5 mph, and Oklahoma City).

where a1, a2, and a3 are regression constants depending on AC

thickness, PG, traffic speed, VFB, and environmental location.
The coefficient of determination (R2) for all 756 polynomial the inuence of AC layer thickness on fatigue damage. This gure
relationships representing the whole database (4,536 computer shows that the inuence of the AC layer thickness on fatigue damage
simulation runs) was always higher than 0.99, indicating excellent is related to the Ecf value. A signicant reduction in the amount of
model predictions. fatigue damage occurs with increased AC layer thickness, especially
when the thickness is more than 3 or 4 in. When a pavement section
has less than that thickness, the damage reduction with AC thickness
Binder Grade shows a quite complicated trend. Depending on foundation stiff-
ness, the fatigue damage either decreases or increases or both with
Analysis of the results also showed that binder stiffness affects the
increasing AC thickness. This result led to a decision to develop the
development of fatigue damage and should be considered as a variable
fatigue damage model with two different sets of coefficients: one for
in the nal fatigue damage model.
less than 3 in. of AC thickness and the other for more than 3 in.

Voids Filled with Bitumen

Effective Asphalt Concrete Dynamic Modulus
The VFB was found to be an important factor, signicantly affecting Predictive Model Development
development of fatigue damage.
The effective AC dynamic modulus (E *eff ) is dened as the single
dynamic modulus at which the amount of predicted bottom-up fatigue
Traffic Speed damage equals the bottom-up fatigue damage that would occur with
variable (seasonal) dynamic modulus values, which change as tem-
Simulation results showed a signicant change in fatigue damage perature and frequency change. To develop an E *eff predictive method-
with the change in traffic speed. A dramatic decrease in fatigue ology, the 1999 version of the Witczak dynamic modulus predictive
damage occurred with increased vehicle speed from creep (0.5 mph) equation was used. This equation has the following form (12, 13):
to slow (15 mph). However, the inuence of the speed change after
log E * = 1.249937 + 0.029232 i p200 0.001767 i ( p200 )
2
15 mph on fatigue damage is not dramatic. A linear relationship was
found between fatigue damage and traffic speed in loglog space. This
relationship can be expressed as follows: Vb eff
0.002841 i p4 0.058097 i Va 0.8022 i
(Vb eff + Va )
D = a b (10)
3.87197 0.0021 i p + 0.003958 i p
4 38
where is traffic speed (mph) and a and b are regression coefficients 0.000017 i ( p38 ) + 0.00547 i p34
2

depending on AC layer thickness, AC mix characteristics, PG, + i log( f ) 0.393532 i log( )) (11)
1 + e( 0.60331300.31335
composite foundation moduli, and environmental location.
where

AC Thickness E* = asphalt mix dynamic modulus, 105 psi;

= bitumen viscosity, 106 poise (at any temperature);
AC thickness is one of the most important factors that signicantly f = loading frequency, Hz;
affects alligator fatigue damage. Figure 1 is an example plot showing Va = air voids in the mix, by volume, %;
El-Badawy, Jeong, and El-Basyouny 119

Vb eff = effective bitumen content, by volume, %; Effective Temperature for Fatigue Cracking
p34 = amount retained on 34-in. sieve, %;
p38 = amount retained on 38-in. sieve, %; The effective temperature (Teff) is dened as the single temperature
p4 = amount retained on No. 4 sieve, %; and at which the amount of predicted bottom-up fatigue damage for a
p200 = amount passing No. 200 sieve, %. given pavement structure is identical (equivalent) to the bottom-up
fatigue damage that would occur from the seasonal uctuation of
This equation clearly shows that E * is a function of VFB, aggregate temperature under cumulative damage principles throughout the
gradation, and frequency and binder stiffness. The binder stiffness, annual environmental cycle.
in turn, depends on the temperature, as shown in Equation 12: The effective temperature for alligator fatigue damage was found
to be a function of the effective frequency and the climatic factor
log log = A + VTS log T (12) as follows:

where
Teff = 2.3316 feff 13.9951 + T f (15)
= binder viscosity, cP;
T = temperature in Rankine scale, R (R = F + 459.67);
and or
A and VTS = constants.
Equations 11 and 12 show that, to calculate E *eff, it is important Teff = 2.3316 feff 13.9951 + 1.0056 ( MAAT ) + 0.8755 ( MMAT )
1.1861( wind ) + 0.5489 ( sun ) + 0.0706 ( rain )
rst to develop a methodology to calculate effective frequency and
(16)
effective temperature.

Details of the development of this equation are shown later. The

Effective Frequency climatic factor (T f ) in Equation 15 is exactly the same form as the
climatic factor used for the AC rutting effective temperature (7, 10).
Figure 2 shows a cross section of the typical pavement structure and Finally, the effective AC modulus can be calculated with the
the traffic loads used to calculate the time of loading. The time of help of the 1999 Witczak predictive model (Equation 11) by using
loading is necessary for calculating the effective loading frequency, the effective frequency and the binder viscosity at the effective
which is necessary for predicting the effective temperature and temperature (12).
effective modulus of the asphalt mixture. The loading time and the
effective frequency were calculated at the bottom of the AC layer,
which was chosen because that is where the maximum tensile strain Fatigue Damage Predictive Model Development
occurs. The time of loading (t) can then be calculated according to
Figure 2, as follows: A comprehensive analysis of the output data from the latest MEPDG
(Version 1.0) results showed that several key variables need to be
2 ( a + hac ) considered in developing the nal fatigue damage model: traffic speed,
t= (13) ESAL, VFB, climatic factor (Tf ), composite foundation modulus (Ecf ),
17.6
AC thickness, and binder stiffness. The AC modulus itself depends
heavily on VFB, traffic speed (frequency of loading), binder viscosity,
where a is the radius of tire pressure (4.886 inch) and is vehicle
and climate (temperature). Thus, fatigue damage can be related to
speed (mph).
the AC modulus.
Finally, the effective loading frequency ( feff) can be calculated as
On the basis of the simulated 4,536 runs, a general comprehensive
the reciprocal of time:
model to predict fatigue damage as a function of the effective AC
modulus, AC thickness, VFB, and Ecf was developed. The E eff * depends
1
feff = (14) on the aggregate gradation of the AC mixture, VFB, traffic speed
t (effective loading frequency), and binder viscosity at the effective
temperature.
As mentioned previously, two sets of coefficients were developed
P = 9000 lb and the selection of either coefficient set is determined by AC thick-
= 120 psi ness of the pavement structure of concern. It is true, in general, that
fatigue damage is reduced as AC thickness increases or sub-AC layer
(i.e., composite foundation) stiffness becomes stronger. However, if
a pavement structure has thin AC thickness along with a very strong
2a
1:1 1:1 composite foundation, little or no amount of tensile strain at the
hac bottom of the pavement structure will be induced by loading, or
compressive strain may be induced in the pavement. That is, predicted
fatigue damage would be zero or close to zero. In this case, as the
2 a + 2 hac
AC thickness increases, tensile strain (i.e., fatigue damage) will
FIGURE 2 Pavement structure for calculating the time of loading increase up to a point and then begin to decrease. For the completed
(P loading, tire pressure). sensitivity analysis, it was found that the transition thickness was
120 Transportation Research Record 2095

placed at 34 in. Therefore, the nal suggested model is determined Nonlinear optimization was performed simultaneously on Equa-
as follows: tions 16 and 17 to determine the regression coefficients for the Teff
model as well as the general fatigue damage model. Once Nf is deter-
For hac 3 in. (thin model): mined, the fatigue damage at the ESALs of interest can be calculated
by using Equation 1. In addition, fatigue cracking can be calculated
{(
log Nf = 8.3014 0.0095 log ( hac ) 0.0756 log ( hac )
2 from predicted damage by using the MEPDG transfer function shown
in Equation 6.
+ 0.0438 ) log ( E eff* ) 0.5414 log ( hac ) + 1.4319 log ( hac ) Figure 3 compares damage predicted with the developed general
2

model and damage predicted with MEPDG for the 4,536 computer
1.0252 log ( Ecf ) + 0.0208 log ( E eff
* ) + 0.7040 log ( E eff
*)
2 2
simulation runs. The model shows excellent prediction accuracy with
Se/Sy = 0.045 and adjusted R2 of 0.998 on the logarithmic scale.
4.1171 log ( Ecf ) + 4.1659 log ( hac ) 3.0733 log ( hac )
2
A discontinuity problem has arisen with regard to the presence of
two sets of coefficients in the developed model because, in reality,
6.4418 log ( VFB) + 1.5883 log ( hac ) 2.8014 log ( hac ) the model produces two predicted fatigue damage values for 3-in.
2 2

AC layer pavement. This problem would be of more concern for

9.2885 log ( VFB) 0.1177 log ( E eff
* ) + 0.0681( hac ) probabilistic analysis where standard deviation of AC thickness is
2 2

taken into account.

0.3789 ( hac ) + 0.8989 log ( E eff
* ) + 2.9330 } (17a) To mitigate the discontinuity, a transition range of AC thickness
was suggested. When the AC layer is somewhere between 2.5 and
3.5 in., fatigue cracking can be calculated with the following equation:
For hac 3 in. (thick model):
fatigue cracking = ( fatigue cracking predicted by thick model )
{(
log Nf = 8.3014 0.0645 log ( hac ) 0.0144 log ( hac )
2

+ (1 ) ( fatigue cracking predicted by thin model ) (18)

+ 0.0416 ) log ( E *eff ) 0.6003 log ( hac ) + 0.7046 log ( hac )
2
where is 2.5 + hac.
When AC thicknesses are 2.5 and 3.5 in. (i.e., = 0 and 1),
1.0276 log ( Ecf ) + 0.0218 log ( E eff
* ) + 0.6280 log ( E eff
*)
2 2

Equation 18 simply becomes the original thin and thick model,

3.2499 log ( Ecf ) + 28.9186 log ( hac ) 51.9588 log ( hac )
2 respectively.

+ 12.7671 log ( VFB) + 15.8844 log ( hac ) 28.6128 log ( hac )

2 2
Composite Foundation Modulus
and Section Transformation
+ 0.9160 log ( VFB) 0.1792 log ( E eff
* ) + 0.0024 ( hac )
2 2

The proposed fatigue damage model was developed based on a two-
0.1009 ( hac ) + 1.2623 log ( E e*ff ) + 1.4613 } (17b) layer pavement system, assuming that the foundation used in the study

4
Log Fatigue Damage (%) from Proposed Model

y = 0.998x
2 R = 0.998

-2

-4

-6

-8
-8 -6 -4 -2 0 2 4 6
MEPDG Ver. 1.0 Log Fatigue Damage (%)

FIGURE 3 Fatigue damage predicted with MEPDG versus damage predicted by using proposed
fatigue model (K degrees of freedom, SES sum-of-squared error, Se standard error
of estimate, Sy standard deviation of criterion variable, and Se/Sy standard error ratio).
El-Badawy, Jeong, and El-Basyouny 121

was a combined layer that represented all layers below an AC layer. A study to nd out the best case (most accurate) among them was
Because pavement systems with three or more layers are generally the conducted by comparing the maximum tensile strain of a four-layer
rule rather than the exception, the need to convert multilayer moduli system with that of a transformed three-layer system using each case.
below the AC layer into a composite foundation modulus (Ecf ) came It was necessary to compute the tensile strain for various pavement
to the front in order to apply the developed fatigue predictive model conditions with the elastic layer program JULEA. As a result of the
to all possible pavement layer systems. study, it was observed that the tensile strains between the two systems
The fundamental theoretical concept to develop the conversion are very close to each other when using Case II where the relevant
method is that the same or very close maximum tensile strain at the equation is as follows:
bottom of the AC layer must occur in both two-layer and multilayer
systems and the modulus in the two-layer system would be the Ecf.
he = ( h3 )
3 E3
To find the Ecf with which the same maximum strain occurs in the (19)
E2
multilayer system, it was realized that calculation of the tensile strains
in the multilayer system is critical.
where
Transformation from a three-layer to a two-layer system was
studied rst. Many combinations of layer conditions having a prob- he = equivalent (combined) thickness for base and subbase
able modulus and layer thickness range for each layer were considered. layers,
The selection of typical pavement layer moduli and thicknesses h3 = subbase thickness, and
in a three-layer system was based on common ranges normally E2 and E3 = base and subbase moduli.
encountered in practice. The elastic layer program JULEA was used
to predict the maximum tensile strain at the bottom of the AC layer for
the three-layer and two-layer pavement systems. Pavement conditions VALIDATING THE DEVELOPED FATIGUE
with various AC and base layer thicknesses were considered to CRACKING PREDICTION METHODOLOGY
make a factorial matrix for both systems, which contributes to a total
of 5,352 JULEA simulations. Thus far, the whole process to predict fatigue cracking by using the
To capture the maximum tensile strain at the bottom of the AC developed predictive models has been established. It includes trans-
layer, the tensile strain was determined at the 10 different locations forming the multilayer system into a two-layer system, calculating
MEPDG uses in the tensile strain calculations (5). The 18-kip standard an effective AC modulus, and nally predicting the load-associated
axle load properties were used for the calculations. The Ecf values fatigue damagecracking based on the effective modulus. The next
were then estimated by using the relationships between Ecf and tensile step was to validate the process by comparing outputs obtained from
strain in the two-layer system. MEPDG and the developed prediction methodology for a three-layer
The tensile strain and Ecf have a three-order polynomial relationship pavement system.
on a logarithmic scale. By developing similar plots at various AC layer A total of 1,000 simulations running both MEPDG and the
conditions, Ecf values in all cases were estimated. All estimated Ecf developed methodology have been completed to validate the process.
values were tabulated according to the five parameters used in the The simulations embrace various conditions that could be encountered
analysis matrix: AC thickness, AC modulus, base thickness, base in the eld. The condition varies by the following parameters: binder
modulus, and subgrade modulus. The table is used to nd a composite grade, Va and Vb eff, AC thickness, AC modulus, traffic speed, base
foundation modulus to convert a three-layer pavement system into a thickness, base modulus, subgrade modulus, and environmental
two-layer system. If Ecf is placed somewhere between the values used location. All these simulation runs were performed at traffic repetitions
in the analysis, the Ecf value would be obtained by using a simple of 2 million ESALs. Table 2 summarizes the number of simulations
interpolation method. depending on each variable used in the validation study.
Another transformation method was introduced in that a pavement Figure 4 compares percentage fatigue damages predicted from both
system having AC layer, base, subbase, and subgrade (i.e., a typical applications. Data points simulated results from various combinations
four-layer pavement system) can become a three-layer system having of environmental and physical road conditions; variables contain not
AC layer, combined base layer, and subgrade. The fundamental only particular values used in the damage predictive model and the
concept used to develop this conversion method is similar to what layer transformation but also values between the initially considered
was used to nd the Ecf. The concept is that the same or a very close ones. More than 70% of the simulation runs matrix used for the vali-
maximum tensile strain at the bottom of the AC layer must occur dation process of the developed alligator fatigue damage methodology
in both three-layer and four-layer systems. A common and widely contains levels for each variable considered in the developed models
known method to combine two layers into one layer is to use what is that differ from the levels initially used in developing the predictive
called Odemarks transformation. It is based on the theory of elastic- methodology. For example, the fatigue damage model was devel-
ity with the assumption that any stresses and strains occurring below oped based on three climatic locations; for the validation process,
a layer depend only on the stiffness of the layer (14). This simple 10 climatic locations distributed all over the United States were used.
concept makes it possible to have either an equivalent thickness or an Figure 4 shows very good agreement between the percentage
equivalent moduluswhichever represents a combined characteristic alligator fatigue damage predicted with the developed methodology
of the separate basesubbase layers. and MEPDG. The adjusted R2 is 0.982 and Se/Sy is 0.134 on the
Three case studies of the Odemarks method were investigated to logarithmic scale.
choose the most accurate case. In Case I, the conversion is to use the
sum of the base and subbase thicknesses as a combined layer thickness
and the moduli of each layer are changed into an equivalent modulus. OVERALL PREDICTION ACCURACY
In Case II, the conversion is to use the base modulus as a combined
layer modulus and only the subbase thickness is changed. In Case III, Because all MEPDG simulations were run at only one traffic rep-
the conversion is to use the subbase modulus as a combined modulus etition of 2 million ESALs, an attempt was made to calculate the
and only the base thickness is changed. damage values at various traffic levels of 5, 10, and 50 million
122 Transportation Research Record 2095

TABLE 2 Summary of MEPDG Simulation Runs Used for Validation of Alligator Fatigue Cracking Methodology

PG No. Va (%), No. AC Thickness No. Environmental No. Traffic No.

Grade of Runs Vb (%) of Runs (in.) of Runs Location of Runs Speed (mph) of Runs

52-40 80 4, 16 120 1 70 Chicago, Ill. 100 0.5 140

58-28 350 6, 14 420 1.5 40 Great Falls, Mont. 100 45 310
64-22 110 8, 12 120 2 70 Homer, Alaska 100 55 310
76-16 340 10, 10 270 2.5 70 Houston, Tex. 100 65 240
82-10 120 12, 8 70 3 100 Indianapolis, Ind. 100
3.5 100 Jackson, Miss. 100
4 190 Oklahoma City, Okla. 100
5 100 Phoenix (Deer), Ariz. 100
5.5 30 Seattle, Wash. 100
6 100 Washington, D.C. 100
7 100
8 30
Total 1,000 Total 1,000 Total 1,000 Total 1,000 Total 1,000

ESALs for ensuring the developed models overall prediction is sensitive to AC thickness and becomes more sensitive with
accuracy. Predicted damage values were then transformed into decreased AC thickness. By simply excluding the thin sections, it
fatigue cracking by using the calibrated MEPDG transfer function can be observed that the scattered pointes were eliminated as
(Equation 6). shown in Figure 5. Therefore, it is obvious that the model would
Fatigue cracking values predicted from the proposed fatigue play more accurately when applied for AC thickness 2 in.
model at the four traffic levels were compared with MEPDG values
as shown in Figure 5; fatigue cracking values in most cases are ratio-
nally predicted by the proposed model, as indicated in the goodness- SUMMARY AND CONCLUSIONS
of-fit statistics (adjusted R2 value is 0.984 and Se/Sy is 0.127 for
hac 2-in. sections). A comprehensive fatigue damagecracking predictive methodology
Few scattered points were observed for the very thin AC thick- was developed based on MEPDG results. The developed model is
ness (<2.0 in.) because the damage-to-cracking transfer function formed as a nonlinear regression model having structure- and material-

2.5 1.0 in.

Log Fatigue Damage (%) from Proposed Model

1.5 in.
2 2.0 in.
2.5 in.
1.5 3.0 in.
3.5 in.
1
4.0 in.
5.0 in.
0.5
5.5 in.
6.0 in.
0
7.0 in.

-0.5 8.0 in.

-1

-1.5

-2
-2 -1.5 -1 -0.5 0 0.5 1 1.5 2 2.5 3
MEPDG Ver. 1.0 Log Fatigue Damage (%)

FIGURE 4 Percent fatigue damage predicted from MEPDG and developed predictive process
(all AC thicknesses).
El-Badawy, Jeong, and El-Basyouny 123

Base Thickness No. Base Modulus No. Subgrade Modulus No. AC Modulus No.
(in.) of Runs (ksi) of Runs (ksi) of Runs (ksi) of Runs

4 160 15 240 3 160 <500 179

5 560 30 560 5 320 >500 821
6 130 40 80 8 150
8 150 50 70 10 340
80 50 30 30

Total 1,000 Total 1,000 Total 1,000 Total 1,000

related variables: AC thickness, composite foundation modulus, VFB, an important role in the developed model, supporting the fact that the
and dynamic modulus. The accuracy of the model was validated by dynamic modulus test is used as an SPT for load-associated fatigue
using randomly selected pavement sections under a variety of struc- cracking.
ture, material, and climatic conditions. Statistics verified that the A major advantage of the developed approach is that the stochastic
developed model reasonably reproduced the amount of fatigue distress analysis can be conveniently performed in a matter of minutes for a
quite close to that from MEPDG. given construction problem, compared with multiday solutions when
The developed methodology incorporates an effective temperature the actual MEPDG is used as the fatigue distress prediction tool. This
approach and an effective dynamic modulus. The concept of the situation occurs because the stochastic analysis requires dozens to
effective temperature is useful simply to characterize a climatic effect hundreds of simulationsdepending on the application type used
on a pavement distress at a given location by one representative to obtain an average and a standard deviation. Implementation of this
temperature value. In addition, the effective dynamic modulus that advantage into the stochastic analysis is quite signicant with regard
comes from a dynamic modulus with an effective frequency plays to the QA practice because PRS in the QA is becoming more common.

100

90
2.00E+06 ESAL
5.00E+06 ESAL
Fatigue Cracking (%) from Proposed Model

80
1.00E+07 ESAL
5.00E+07 ESAL
70

60

50

40

30

20

10

0
0.0 10.0 20.0 30.0 40.0 50.0 60.0 70.0 80.0 90.0 100.0
Fatigue Cracking (%) from MEPDG, hac 2 in.

FIGURE 5 Fatigue cracking at various traffic levels between MEPDG and proposed model
(AC thickness 2 in.).
124 Transportation Research Record 2095

The QA practice is powerful when supported by the statistically 5. ARA, Inc., ERES Consultants Division. Guide for MechanisticEmpirical
based PRS. Design of New and Rehabilitated Pavement Structures. Final report,
NCHRP Project 1-37A. Transportation Research Board of the National
Thus, it is recommended that the developed methodology be
Academies, Washington, D.C., 2004. www.trb.org/mepdg/guide.htm.
implemented as one of the performance predicted models for NCHRP 6. Witczak, M. W., K. E. Kaloush, T. Pellinen, M. El-Basyouny, and H. Von
Project 9-22. Quintus. NCHRP Report 465: Simple Performance Test for Superpave
Mix Design. Transportation Research Board of the National Academies,
Washington, D.C., 2002.
7. El-Basyouny, M. Calibration and Validation of Asphalt Concrete
ACKNOWLEDGMENTS
Pavements Distress Models for 2002 Design Guide. PhD dissertation,
Arizona State University, Tempe, 2004.
The authors acknowledge that this study was conducted under the 8. El-Basyouny, M., and M. W. Witczak. Verication for the Calibrated
general supervision of M. W. Witczak. The work was conducted as Fatigue Cracking Model for the 2002 Design Guide. Journal of the Asso-
a part of NCHRP 9-22 (Beta Testing and Validation of HMA PRS). ciation of Asphalt Paving Technologists, Vol. 74, 2005, pp. 653695.
Fugro Consultant Inc. is the principal agency and James Moulthrop 9. Sotil, A. Use of the Dynamic Modulus E* Test as Permanent Deformation
performance Criteria for Asphalt Pavement Systems. PhD dissertation.
is the project principal investigator.
Arizona State University, Tempe, 2005.
10. Witczak, M. W., and A. Sotil. Determination of The Effective Tempera-
ture of Asphalt Mixtures for Permanent Deformation Analysis. Interteam
REFERENCES Technical Report, NCHRP Project 9-19. Arizona State University, Tempe,
2004.
1. Witczak, M., A. Patni, and S. El-Badawy. Development of a Rational 11. El-Basyouny, M., and M. G. Jeong. Effective Temperature for Permanent
Methodology to assess Performance Related Pay Factors for Asphalt Deformation and Fatigue Distress Analysis on Asphalt Mixtures. Pre-
Pavements. ASU-ADOT Research Project, Arizona State University, sented at 88th Annual Meeting of the Transportation Research Board,
Tempe, 2006. Washington, D.C., 2009.
2. Witczak, M., and S. El-Badawy. Development of a Spreadsheet Solution 12. Witczak, M. W., D. Andrei, and W. Mirza. Development of Revised Pre-
to Evaluate AC Mixtures Using E* as a SPT for Fatigue. NCHRP 9-22 dictive Model for the Dynamic (Complex) Modulus of Asphalt Mixtures.
Project Interim Status Report No. 1. Arizona State University, Tempe, Interim technical report, NCHRP Project 1-37A. University of Maryland,
2005. College Park, 1999.
3. Witczak, M., S. El-Badawy, and A. Patni. Development of a Spread- 13. Bari, J., and M. W. Witczak. Development of a New Revised Version
sheet Solution to Evaluate AC Mixtures Using E* as a SPT for Fatigue. of the Witczak E* Predictive Models for Hot Mix Asphalt Mixtures.
NCHRP 9-22 Project Interim Status Report No. 2. Arizona State Uni- Journal of the Association of Asphalt Paving Technologists, Vol. 75,
versity, Tempe, 2005. 2006, pp. 381417.
4. Witczak, M., S. El-Badawy, and M. G. Jeong. Development of a Rational 14. Ullidtz, P. Modeling Flexible Pavement Response and Performance.
Methodology to Access Performance Related Pay Factors for Load Polyteknisk Forlag, Lyngby, Denmark, 1998.
Associated Alligator Fatigue Cracking. NCHRP 9-22 Project Interim
Status Report No. 3. Arizona State University, Tempe, 2006. The Flexible Pavement Design Committee sponsored publication of this paper.