A Model for Pavement Design and Management Based on the

AASHTO Guide and HDM Regressions

Paper submitted to the Transportation Research Record
Prof. Rgis Martins Rodrigues (Aeronautics Technological Institute, ITA, Brazil)

ABSTRACT

This paper describes the technical base of the performance prediction model that is at the core
of a pavement management system developed for Brazilian highways, which main feature is
to reduce to a minimum a major shortcoming of existing systems: the usually great
discrepancy between maintenance solutions and costs established for planning purposes at
network level and the ones that are effectively detected after studies are made at the project
level. This was accomplished through close attention to the pavement structure-traffic
interaction feature of the performance prediction model that is employed in the system. For
this purpose, a generalization of the 1986 AASHTO Guide model was developed, introducing
long-term pavement performance effects through the use of World Bank HDM-III empirical
models. Application of the model to a set of independent pavement performance cases gave a
strong support for its reliability. It was also possible to extend the applicability of this model
to airport pavements, since its results are in good agreement with FAA design procedures and
USACE models.

Key words:
Highway pavements airport pavements pavement design and management

1 - Introduction
At network level, when a Pavement Management System (PMS) applies performance
prediction models, they are usually of a simplistic nature, due to the requirement of keeping
the monitoring of field data at low costs per kilometer. Besides, several systems attempt only
to detect, at one side, the pavement segments where minor maintenance measures can be
applied and, at the other side, the ones where a major rehabilitation should be scheduled. In

these versions of a PMS there is no need for truly reliable performance prediction models at
network level, but the usefulness of such systems is severely restricted to prioritization of
maintenance interventions at a moment when filed monitoring data are available. Nothing can
be done about multi-year planning and budget analyses. For these, remaining life estimates are
a critical requirement.
On the other hand, experience with a PMS applied to several highway networks in
Brazil led to the need for performance models that go further: they should have the capability
to anticipate structural needs that will not be grossly over- or underestimated when
rehabilitation designs are performed at project level. A complete agreement between network
and project level solutions is impossible, since additional factors must be taken into account at
project level, such as localized environmental (climatic, drainage, pavement geometry) and
traffic conditions, pavement history and detailed pavement condition and layers materials
mechanical properties. However, when managerial characteristics impose the need for a good
agreement between these two levels of decision, the performance models to be applied at
network level must have adequate accuracy to deal properly with structural parameters (traffic
loads and their interaction with the pavement structure). This is particularly important in
Brazil due to the broad range covered by the bearing capacities of highway pavements, with
major extensions of highways operating, due to budget constraints, with relatively thin
pavements under heavy traffic loading conditions.
In fact, applications of the World Bank HDM system in Brazil have been deemed to
failure due to misconceptions on its use. Instead of running the program only for pre-selected
strategies, generated after sound technical considerations have been made regarding pavement
performance, strategies were implemented based solely on least initial cost and the resulting
expected benefits due to vehicle operation costs decrease. This led, for example, to thin
resurfacings applied over severely cracked pavements without anti-reflection cracking interlayers, or to thin overlays applied to structures experiencing bad performance due to excessive
plastic strains developing in the layers below the surfacing layer. Besides, lack of a proper
consideration regarding traffic-structure interaction by the empirical correlations on HDM led
to false expectations about rehabilitated pavement service lives under the use of these thin
overlays. Such experience highlights the importance of a PMS with sound performance
prediction models and reliable decision trees for the detections of pavement maintenance
needs.
The PMS here mentioned has been in use for over 14 years, mainly for the following
purposes:
2

(a) For budged estimates in the near future, and sometimes at longer analysis periods;
(b) As an aid in rehabilitation designs at project level, furnishing diagnoses, maintenance
solutions and pavement and traffic data;
(c) On special studies, so that ones involving simulations of different performance
patterns required for the pavements on the network and the costs associated with the
attainment of these patterns;
(d) As a tool for designs made before an actual design at the project level is conducted, so
that a fast and reasonable estimate of rehabilitation needs can be obtained, as well as to
avoid gross mistakes at project level.

Pavement construction and its subsequent maintenance involve consumption of precious

natural resources and of energy. To avoid a wasteful use of these resources is a central task for
any PMS, which requires truly reliable pavement performance prediction models in order to
optimize the allocation of limited funds to the purpose of maintaining an adequate service
level for the highway users.
The final purpose of a PMS is the discovery of the most effective pavement
maintenance strategy, for which a measure of the relative efficacy of different strategies is
required. A parameter to be maximized on a pavement construction or rehabilitation strategy
is the Benefit-cost Ratio (BCR) associated with the investment, which can be defined as:

BCR =

PSI av
C eq
[1]

where PSIav is the average pavement serviceability and Ceq is the equivalent uniform annual
maintenance cost, associated with the investment during its service life (Lf), or:

PSI av
C eq =

1
=
Lf

Lf

PSI (t )dt
0

COST
Lf

[2]
Such a definition for BCR allows the comparison to be made of different design periods
(Lf) and different standard levels for pavement performance (PSI0 and PSIt on Figure 1) so
that the most efficient solution can be found.
3

The achievement of a certain pavement performance is not, however, a certainty, due

to the stochastic nature of pavement behavior. Each design has its proper reliability level (RL),
which represents the probability of the outcome expressed by PSIav. Therefore, the best
solution is the one that maximizes the Expected Utility, defined by:

EU = p r U = RL BCR = RL

PSI av
C eq
[3]

The maximization of this parameter involves the application of reliable models for the
prediction of pavement performance in relation to its reliability level.

PSI
PSI0

PSIt

Lf

Figure 1 Serviceability with time for a pavement investment

2 Development of the Performance Prediction Models

Based on the previous discussion, the following characteristics were judged useful for the
performance prediction models:

(1) The models must be able to describe the long-term performance of highway
pavements;
(2) The interaction between pavement structure and traffic loads must be accurately dealt
with;
(3) Inputs for the model should be the data usually available from network monitoring
programs;

(4) The models should be easily calibrated, so that remaining life estimates could be
made using pavement past performance;
(5) Pavement condition should be adequately and easily described, as well as its
modification through pavement maintenance or rehabilitation;
(6) Due consideration must be given to the stochastic nature of pavement performance, so
that a desired reliability level for the performance predictions can be established as a
fundamental parameter for the conception of a maintenance strategy.

Existing performance prediction models aimed at pavement management applications,

such as the HDM models (Paterson, 1987), have limited applicability due to their empirical
nature and their close association with a limited set of field data. They have, however, the
advantage of being able to describe long-term pavement performance. On the other side, a
design model such as the AASHTO Design Guide (1986 or 1993) is strong at describing
traffic-structure interaction, but is not compatible to long-term pavement performance effects,
due to the accelerated nature of the AASHO Road Test (1958-1960), since layers material
properties change continuously with pavement age, so that strength parameters may increase
by 50% ten years after construction (Croney and Croney, 1991). Therefore, the procedure here
adopted for the development of the performance prediction models was to associate these two
powerful models, avoiding their shortcomings and using their strengths.
The present condition of a pavement can be described in a comprehensive manner by
the use of three indicators:

(a) A serviceability parameter, related to user riding comfort and vehicle operation costs;
(b) A pavement distress index, useful as a gauge to anticipate pavement maintenance
needs and to keep the risks of a sudden pavement failure under control;
(c) Security related indexes, so that the ones related to directional control (hydroplaning
potential), breaking distance (friction on wet surfaces) and obstacles (potholes,
surfacing disintegration).

The third set of indicators should be addressed through proper network monitoring of
the pavement surface condition, whereas the first two are affected by traffic-structure
interaction, so that they will be the ones addressed in this paper for development of
performance prediction models.

The first parameter can be the AASHTO Present Serviceability Index (PSI). Use of the
PSI has the advantages of being linked to a sound model describing the traffic-structure
interaction, present at the 1986 or 1993 AASHTO Guide for Design of Pavement Structures,
and of being in a close relationship with riding comfort, as expressed by:
QI

71.5

PSI = 5e

[4]
where QI is the Quarter-Car Index, in counts per km, that measures longitudinal roughness
(QI = 13IRI, where IRI is the International Roughness Index, m/km). This correlation was
derived in a World Bank experimental study (Paterson, 1987).
The second parameter should be a function of the nature, extent and severity of
pavement distresses, as given by a Surface Condition Index (SCI), which will be here defined
on the same scale of the PSI, from 0 to 5, and with the same meaning regarding their
respective levels, as shown in Table 1.

Table 1 Scale for the PSI and the SCI

Range
45
34
23
12
01

Pavement Condition
Excellent
Good
Regular
Bad
Very bad / Pavement failure

For this purpose, the SCI was determined from a relationship with a global distress
index known as the IGG, Global Gravity Index, (Pereira, 1979), given by:

SCI =

309,22 0,616 IGG

61,844 + IGG
n

IGG = f i pi
i =1

[5]
The IGG is thus determined from the nature, extent and severity of the existing surface
distresses, as the summation of the products of the corresponding frequency (fi) versus
weighing factor (pi) of the n existing defects. Table 2 shows these weighing factors as used in

Brazil. A close relationship exists between this parameter and the USACE Pavement
Condition Index: PCI = -2.7E-6 * (IGG ^ 3) + 0.0025 * (IGG ^ 2) - 0.8109 * IGG + 100.

Table 2 Weighing factors for the IGG (Pereira, 1979)

Distress Type
Low severity cracks
Raveling
Average Rutting
Medium severity cracks
Rutting Variance
Bleeding
Repairs
High severity cracks
Depressions
Potholes and Corrugation

fi pi
fi 0.2
fi 0.3
1.33 RD mm ( 30 mm)
fi 0.5
Variance ( 50 mm2)
fi 0.5
fi 0.6
fi 0.8
fi 0.9
fi 1

Code
FC-1, TRR
D
RD
FC-2, TB
RDV
EX
R
FC-3, TBE
ALP, ATP
P, O

A model for roughness progression can be written in terms of traffic, expressed as the
number of accumulated standard 80 kN axle loads N until pavement age t is reached, as:
IRI( N ) = IRI 0 e N
[6]
where is a function of the pavement structural strength and of the environmental conditions
(climate, geometry, drainage). In this model, the product N expresses the traffic-structure
interaction, which controls the increase of IRI with time. IRI0 is an intercept at pavement age
equal to zero and is nearly equal to the initial pavement roughness. This functional shape is
supported by several pavement performance data (see, for example, Figure 2), especially
during periods of time when IRI > 0.5 m/km. For the initial years after construction or
rehabilitation, a linear rate of growth establishes after the short term consolidation, but an
exponential rate of growth is the norm as pavement deterioration advances with pavement
age, so that the function of equation [6], with only two parameters, is adequate for practical
purposes.
Merging equations [4] and [6] leads to:

PSI 0
PSI ( N ) = 5

exp ( PSI N )

[7]

IRI (m/km)

5
y = 0.6231e0.0853x
R2 = 0.9962

2
15

17

19

21

23

25

Age (years)

Section SHRP_ID = 3005, STATE_CODE = 6

1.25
y = 0.798e0.0225x
R2 = 0.8404

IRI (m/km)

1.20
1.15
1.10
1.05
1.00
11

12

13

14

15

16

17

18

19

Age (years)

Section SHRP_ID = 1001, STATE_CODE = 4

3.5
y = 0.0682e0.2524x
R2 = 0.9258

IRI (m/km)

3.0
2.5
2.0
1.5
1.0
11

12

13

14

15

Age (years)

Section SHRP_ID = 1002, STATE_CODE = 25

Figure 2 Sections from LTTP DataPave 3.0
8

with PSI0 being the initial pavement serviceability index. The functional shape of equation [7]
has all the desirable features surrounding parameter PSI evolution with time, as revealed by
long-term pavement performance data, a critical feature on any reliable performance
prediction model (Lytton, 1987). Similarly, a hypothesis will be launched here involving the
practical applicability of writing:

SCI 0
SCI ( N ) = 5

exp ( SCI N )

[8]
where SCI0 is the initial (new or newly rehabilitated pavement) Surface Condition Index. Up
to this point, nothing can prove the validity of equation [8], but as will be shown here its
calibration to several pavement data bases has been possible with an acceptable degree of
accuracy.
Besides, this functional shape has the advantage of being associated with a recursiveincremental type of performance prediction model, since it can be rewritten, in differential
form, as:

PSI

PSI = PSI N PSI ln

[9]
The rate of change of the PSI with traffic (PSI/N) is entirely a function of the
present condition (PSI) and of the structural parameter (PSI). This has a clear convenience on
PMSs, since the present condition can be updated from field monitoring data, increasing the
performance prediction reliability for subsequent years.
In equations [7] and [8], if the structural parameters PSI and SCI are known, then
future performance of the pavement can be predicted. In this respect, emphasis must be given
to the determination of how these values vary with pavement structural parameters.
The AASHTO Design Guide (1986) model can be applied in the context of
determining parameter PSI, since it was derived from a test especially designed to evaluate
the traffic-structure interaction.

The AASHTO model is not, however, reliable as a

performance prediction model, due to the accelerated nature of the AASHO Road Test, where
long-term effects could not be evaluated. Therefore, this model will be here applied
exclusively with the objective of arriving at a nominal pavement traffic bearing capacity,
leading to a first estimate of the parameter, capable of including in the prediction process
9

the key variables related to pavement structure and traffic loads. Long-term second order
effects will be taken into account after these key effects are included in the model. In this
respect, the AASHTO design model will be here applied in the same context as it is intended
for in The AASHTO Guide (1986): to furnish an estimate of the accumulated traffic that is
required for a pavement global serviceability loss from a new pavement condition (PSI0) until
pavement failure (PSIf), when a rehabilitation, such as a major asphalt concrete overlay,
will be needed. Therefore, the first estimate of the parameter, as given by the AASHTO
Guide, will be:
PSI 0

ln
5
1

A =
ln
W18 2.5
ln 5

[10]
where W18 is the 18 kips single axle load accumulated number of repetitions associated with a
condition change from PSI0 to PSIf = 2.5. According to the 1986-1993 AASHTO Design
Guide, in the case of asphalt pavements:

W18 = 10

Z R S0

= 0.40 +

(SN + 1)
1.0512

1094

9.36

PSI 0 2.5
4 .2 1 .5

MR
3000 psi

2.32

(SN + 1)5.19
[11]

where SN (= aihi) is the pavement Structural Number for the layers above a certain layer,
MR is the roadbed resilient modulus of this layer (MR < 40,000 psi or 280 MPa) and PSI0 is
the initial Present Serviceability Index. ZR and S0 parameters deal with the stochastic nature of
pavement performance and with the desired design reliability level.
With the most important variables already included in the model by this process
(equation [11] for PSI and the AASHTO load equivalency factors for N), it is time now to
deal with long-term effects, environmental conditions and other factors not included through
the use of equations [10] and [11]. This will be done by defining a calibration factor FC
applied to the parameter, or:

PSI = Fc A

[12]

10

A major effort in this research was at establishing the functional dependence of FC on

its controlling variables. The performance prediction model here proposed can be considered
reliable only if patterns can be identified of FC variation with environmental (climatic,
geometric, drainage) and construction (materials, construction quality and construction
processes) conditions.

3 Calibration from HDM Models

The HDM-III models (Paterson, 1987) are good predictors of long-term pavement
performance, as revealed by several studies, such as the one by Rodrigues (2009), who
applied and recalibrated these models using data from 8,000 km of highways in Brazil.
Changes had to be made on some models, such as in the case of prediction of the first cracks
appearance on overlaid pavements. One of HDM models for such prediction is:

TYcr 2 = 2,54 exp (0,0157 H R 0,0141PCR 4 )

[13]
where:
HR = HMAC overlay thickness (mm);
PCR4 = percent area cracked of the old pavement (AASHTO Class 4).
This formula has a serious drawback, as it does not include two central parameters in
the process: traffic and pavement deflections. Due to this, the model that actually is applied on
HDM-III is:
TYcr 2 = 10,8 exp(1,21D0 1,02YE4 D0 )

[14]
where D0 is the pavement deflection under the 80 kN standard axle load (in mm) and YE4 is
the annual traffic rate, in 106 equivalent 80 kN axle loads.
In order to incorporate both these models, it can be written:

TYcr 2 =

(1.0TYcreq2.(13) + 1.5TYcreq2.(14) )
(1 + 1.5)

[15]

11

as revealed by applications of these models to several Brazilian highways. Such model will
allow consideration of the presence of anti-reflective cracking interlayers, since in this case it
will be possible to state PCR4 = 0 for interlayers capable of filling the upper part of the most
severe pavement cracks. These models were developed for the following set of parameters:
50 HR 125 mm
4 Age 16 anos (pavement age at the time of rehabilitation)
0,02 YE4 1,59

For the prediction of the pavement age in years (TYcr2) at the first crack appearance at
the pavement surface on flexible pavements:
YE4

TYcr 2 = 4,21exp 0,139SNC 17,1

SNC 2

[16]
where SNC is the structural number corrected for subgrade bearing capacity:
2
SNC = SN + 3,51log10 CBR 0,85 log10
CBR 1,43

[17]
The HDM models for rutting and raveling prediction were applied directly, without
any corrections. That was also the case for the longitudinal roughness progression (Paterson,
1987):
IRI (t ) = 134e 0.023t SNC 5.0 NE4 + 0.114RD + 0.0066CR + 0.003H P PAT + 0.16VPOT + 0.023IRI (t )t

[18]
where:
IRI(t) = International Roughness Index at year t (m/km);
t = pavement age since construction or last rehabilitation (years);
NE4 = traffic rate during the time interval t, expressed in 80 kN axle loads

repetitions, in millions;
IRI = increase in the pavement longitudinal roughness (m/km) during the time

interval t (years);
RD = increase in rut depth during the time interval t (mm);

12

CR = increase in surface cracking during the time interval t (% of area cracked with

high severity cracks);

PAT = increase in the area with patching, with mean depth or elevation equal to HP

(mm), in percent area of the pavement;

POT = increase in volume of potholes during the time interval t (m3).

The HDM-III models have several shortcomings as true performance prediction

models, especially when considering their narrow experimental base and lack of mechanistic
considerations, both leading to a loss of prediction power. The effects of a change in thickness
of the asphalt concrete surfacing layer are not properly dealt with by such models, since its
increase only slightly increases pavement service life as predicted by them. Therefore, these
models were here applied only for a selected set of reference conditions, chosen to represent
the experimental data base. It is believed that, for such conditions, the models predictions are
not biased and will reflect the life expectancy under those conditions. This procedure is a need
dictated by the purely empirical nature of HDM models and from the fact that they employ
only few data parameters for the consideration of traffic loading and pavement structure
effects on pavement performance. Because of this, the models are too simplistic to reliably
describe reality, leading them to suffer from a fundamental lack of accuracy to consider the
effects of variations of pavement materials mechanical properties and of pavement layers
thicknesses on pavement performance. Therefore, HDM models can be trusted as a source of
reliable long-term pavement performance data only for pavement sections with similar
characteristics of the ones that comprised the data base that was used in the regression
analyses that led to those models. These data refer to pavements situated on Brazilian
highways, which were designed using the DNER Method (DNIT, 2006), a version of the
original USACE-CBR method. This method requires a total pavement thickness in terms of
granular base materials as given by:

7011

HT ( cm) = 9,02 + ( 0,23 log10 N P + 0,05)

234 ,33
CBR

[19]
for the pavement to support NP equivalent 80 kN single axle load repetitions for a subgrade of
bearing capacity described by its CBR value. Besides, Table 3 shows the minimum surfacing
layer thickness required by the method and this thickness must be converted to equivalent

13

granular base layer thickness using a coefficient equal to 2, so that: HT = 2h1 + h2 + h3 for the
total pavement thickness of a section with surfacing (h1), base (h2) and sub-base (h3) layers.
Table 3 Minimum surfacing layer requirement on DNER Method (DNIT, 2006)
NP (USACE load equivalency factors)
106
106 < NP 5 106
5 106 < NP 107
107 < NP 5 107
NP > 5 107

Surfacing Layer
Surface treatments
5 cm of asphalt concrete
7,5 cm of asphalt concrete
10 cm of asphalt concrete
12,5 cm of asphalt concrete

The resilient modulus of the subgrade soil will be given by: E4 = 10 CBR (MPa). For
the granular base layer (crushed stone), the Shell formula was applied for estimating the
resilient modulus E2, as given by:
E2
= 0.2 h20.45
E3
[20]
for the base thickness (h2) in mm. The granular subbase modulus E3 was estimated as the
average value from equation [20], E3 = 0.2h30.45E4, and E3 = 5 CBRSB(MPa).
It was also required that the selected pavement sections met a maximum deflection
under the 80 kN axle load as expressed by: Dmax (0.01 mm) = 10(3.01-0.176logNp), which is also a
part of DNER method for pavement design and evaluation.
These cases were also selected imposing the condition that fatigue life, associated with
percent area cracked TR = 20%, was around (70 to 125%) the pavement service life, defined
by a terminal condition given by PSIt = 2.5 for low traffic conditions, 2.7 for medium ones
and 2.9 for high traffic volumes. This was considered to be a normal requirement to be met by
a well balanced pavement structure. Fatigue life was calculated using equation [16] plus 4
years for a condition given by TR = 20% area cracked. This condition was employed to find
the calibration factor Ff for the following mechanistic-empirical model:

N f = F f Fthick N TAI
f
[21]
where NfTAI is the MS-1 (1982) fatigue equation from The Asphalt Institute and Fthick is equal
to 1 for HMAC surfacing thicknesses h1 7 cm and Fthick = (h1/10)2 for h1 < 7 cm. Dynamic

14

modulus of the asphalt concrete was calculated from T.A.I. dynamic modulus formula for
traffic speeds at 90 km/h and surfacing layer average temperature of 770F, and the maximum
tensile strain under the HMAC layer was given by:

t = a0 + a1

flex
E2

1500

h
E
+ a 2 flex + a 3 2 + a 4 2 + a5 ln R
r
1500

h E 3
R = 1 1
[22]
h2 E 2
ln flex = 9.250563 + 0.4698504h1 4.651531 10 2 h1 ln E1 + 0.3187476 ln Dmax
with:
r2 = 0.894 (coefficient of determination)
SE = 2.684586 10-5 (standard error of estimate)
SE/SY = 32.6% (SY = standard deviation of the dependent variable)
a0 = -3.8853 10-5

a1 = 1.093891

a2 = -9.543262 10-2

a3 = 1.411541 10-5 a4 = 2.545568 10-7 a5 = 3.507885 10-5

E1 = modulus of elasticity of the asphalt concrete surfacing layer (kgf/cm2)
E2 = resilient modulus of the base layer (kgf/cm2)
h1 = surfacing layer thickness (cm)
h2 = base layer thickness (cm)
Dmax = vertical pavement deflection under the 80 kN axle load (10-2 mm)
r = 10.8 cm = radius of the loaded area under one wheel of the 80 kN dual wheel axle
load.

for a total of 2647 data cases generated using FLAPS program (Finite Layer Analysis of
Pavement Structures), which was applied for the combinations:

h1 = 3 6 9 12 15 cm
h2 = 5 10 15 20 30 cm
E1 = 40000 60000 80000 100000 150000 kgf/cm2
E2 = 1500 5000 10000 30000 90000 kgf/cm2
Ef = 500 1000 2000 3500 7000 kgf/cm2 = foundation elastic modulus,
comprised of subbase and subgrade layers.

15

Considering a set of pavement sections that were generated according to this

procedure, Table 4 shows the resulting calibration factors required in order that the model
here proposed could reproduce pavement performance as described through HDM empirical
models. Three values for FC were calculated, each corresponding to the pavement layer with
resilient modulus MR above which the structural number SN was computed for use on
equation [11]. The low variability of the Fcs values indicate that the calculated average are
representative enough of the 16 selected cases to be taken as the calibration factors to be
applied for pavement design and analysis. A value for Fc = 1.89 can be taken, irrespective of
the way the pavement structure is described in the model.
The FC values on Table 4 were determined so that the empirical performance predicted
using HDM-III models could be nearly replicated for the analysis considering the whole
pavement structure supported by the subgrade soil with resilient modulus given by: MR = 10
CBR (MPa). The other two analyses were conducted with a generalization of the MR
parameter to be used on equation [11]:

MR =

E eq + (10 CBR )
2
[23]

for Eeq and MR in MPa, with Eeq being the equivalent elastic modulus of the foundation that
supports the structure with Structural Number SN (subbase and subgrade layers if SN =
a1h1+a2h2, layers below the surfacing one if SN = a1h1). For the subgrade analysis nothing
changes, since MR = Eeq = 10CBR.
The equivalent modulus Eeq(i) of the system comprised of the layers situated from layer
i and below was calculated using the formula for the semi-infinite elastic half space:

E eq (i )

2 2
heq (i )
heq (i )
(1 + i ) pR
1
(1 2 i )
=
+ (1 2 i )1 +

1
R
R
i

2 2

1 + heq (i )

[24]

16

where i is the vertical deflection under the center of a circular applied load (of pressure p and
radius R) at depth heq(i), which is the equivalent thickness of the pavement layers situated
above layer i, of Poissons ratio i. Using Odemarks method:

(
(

heq (i )

)
)

E j 1 i2
= hj
2
j =1
Ei 1 j
i 1

[25]
The vertical deflection on top of the subgrade can be calculated directly from equation
[24], using the soil resilient modulus for Eeq. For the other layers:

i = i +1 + i hi = i +1 + hi

i
Ei

= i +1 +

hi ( v (i +1) + v (i ) )

Ei
2
[26]

where the vertical compressive stress on top of layer i is given by:

v (i )

= p 1
3
2
2


R
1 +

heq (i )


[27]

Table 4 also shows that it was possible to calibrate equation [21] using Ff = 0.57 to
predict the first cracks appearance at the pavement surface, and using Ff = 1.15 for the
condition given by TR = 20%. The selected reference pavement sections have average fatigue
life Vf, associated with TR = 20%, nearly equal to the service life VS, defined in terms of a
terminal condition for the PSIt = 2.7 under medium traffic conditions. Under light traffic
conditions, PSIt was chosen to be 2.5 and Vf/VS 1.3, whereas under heavy traffic conditions:
PSIt = 2.9 and Vf/VS 0.7. This assures consistency between performance in terms of both
parameters, since it is expected that the increase in commercial traffic will lead to more
fatigue cracking.

17

Table 4 Results for PSI prediction on new asphalt pavements

Case
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
Average =
Std.Dev. =
Cv (%) =

Ff
eq. [21]
0.500
0.973
1.544
1.484
1.879
1.189
1.155
0.792
1.758
1.484
1.603
1.591
0.362
0.748
0.680
0.654
1.150
0.484
42.1

Ff for Vf/VS
Nyear
Subgrade
VS
Fc
Fc
Fc
TYcr2
(AASHTO)
CBR
(years)
Subgrade Subbase Base
0.285 1.91
5.0E+05
10
11
1.850
2.297 2.657
0.585 0.90
5.0E+05
5
13
1.630
2.249 2.104
0.707 0.90
1.0E+06
12
6
1.650
1.841 2.238
0.704 0.83
1.0E+06
10
7
1.610
1.793 2.013
0.708 0.86
1.5E+06
10
5
1.450
1.688 1.946
0.681 0.89
7.0E+05
5
10
1.350
2.108 1.844
0.625 0.95
5.0E+05
15
9
1.770
1.522 1.946
0.459 1.05
3.0E+05
15
13
2.220
1.930 2.489
0.814 0.60
1.5E+06
5
8
1.400
2.045 2.001
0.704 0.83
1.0E+06
10
7
1.600
1.863 2.148
0.726 0.86
1.0E+06
13
6
1.700
1.736 2.326
0.690 0.72
1.5E+06
7
7
1.550
2.027 1.402
0.224 2.06
3.0E+05
7
16
1.980
2.505 3.179
0.466 0.95
3.0E+05
5
17
1.710
1.658 0.891
0.405 1.27
3.0E+05
10
13
1.800
1.831 0.899
0.341 1.82
7.0E+05
12
8
1.700
2.026 2.589
0.570 1.089
1.686
1.945 2.042
Average =
0.181 0.442
Std.Dev. =
0.217
0.259 0.600
PP
31.7 40.6
S0 =
0.129
0.133 0.294

Equation [23] indicates that MR, to be used on the AASHTO model (equation [11])
when analyzing the pavement structure above the base or above the subbase layers, should not
be simply the base or the subbase resilient modulus, but an average value between the
equivalent elastic modulus of the system that supports the structure described by SN and the
expected resilient modulus as a function of the layer strength. This last parameter was given
by the relationship that can be applied only to soils (MR = 10 CBR in MPa), but that was
here adopted for the base and subbase analyses due to the fact that the value of FC on Table 4
is nearly constant, irrespective of the way the pavement structure is described, such as
consisting only of the surfacing and base layers supported by the foundation (subbase and
subgrade layers), the surfacing layer over the system comprised of the layers below it, or as
the complete structure supported by the subgrade. Such result indicates that a granular base or
subbase layer has much more value in the pavement structure due to its shearing strength than
due to its resilient modulus. There was a loss of prediction accuracy for base and subbase
analysis (but not for the subgrade analysis) when the AASHTO design formula was expressed
in terms of the resilient modulus, instead of in terms of a soil support value, more associated
with shearing strength. Equation [23] is a generalization which corrects this problem.
The calibration factor FC is, therefore, nearly equal to 1.89, irrespective of the way the
pavement structure is described, if equation [11] is used with the MR definition of equation
[23]. This result means that use of the AASHTO design equation for the base and subbase
18

analyses must consider MR not as the base or subbase resilient modulus (this can be done only
for the subgrade analysis), but as a function of the foundation equivalent elastic modulus and
of the layer shearing strength, as described by its CBR value.
Application of this model for pavement design or performance prediction requires
determination of the maximum value of = 1.837 A from these three different ways of
describing the pavement structure, since this maximum value for will be the one controlling
pavement performance for a specific pavement structure being analyzed. Besides, such
determination will tell which pavement layer is the critical one that controls design. Such
finding will be an invaluable tool for pavement rehabilitation designs, if reliable data can be
gathered about the in situ effective elastic moduli of the pavement layers, since in this case the
critical layer will be weakest link to be reinforced in order to assure traffic-structure
compatibility.
Table 4 also shows a measure of the uncertainty regarding performance prediction
using the proposed model, in terms of the ratio S0PP = StdDev / Average for FC at the several
cases here analyzed. The AASHTO Guide overall standard deviation parameter S0 has
recommended values around 0.4 - 0.5, which makes sense, since part of this value is due to
the uncertainties solely associated with the act of performance prediction (described by S0PP
on Table 4 from 0.13 to 0.30), mainly associated with pavement layers variation in terms of
their thicknesses and the mechanical properties of the comprising materials. The remaining
part is due to uncertainties from future traffic.
Table 5 and Figure 3 show these calculations for reference Case 2. The procedure here
adopted was to find, by trial and error, the best FC value for the subgrade analysis. The other
two FC values, corresponding to the sub-base and base layers, were simply given by: FC =
/A, with being the value required for pavement performance reproduction. This
procedure forces the model to predict the same output in terms of pavement performance
irrespective of the way in which the pavement structure is described in the model by the pair
of parameters MR and SN.

19

Table 5 Analysis of Case 2 for PSI prediction

Layer
HMAC
Granular Base
Granular Subbase
Subgrade
PP (years) =
HTp (DNER) =
Nyear =
QI0 =
PSI0 QI =
Np (USACE) =
h1_mn. =
VDMc =

h1 (cm)
10
18
40

ai
0.44
0.14
0.11

13
78.1 cm
5.00E+05
17
3.942
1.95E+07
10 cm
913

(Subgrade)
AlphaA =
W18 =
Bheta =
Fc =
Alpha =
(Subbase)
AlphaA =
W18 =
Bheta =
Fc =
(Base)
AlphaA =
W18 =
Bheta =
Fc =

SNC =
TYcr2 =
D0 mm =
Dmax =
Vs (TR=20%) =
Epst_HMAC =
Vf (years, TAI) =
Vf (years) =
Vf / Vs =

5.023
6.03
0.491
53.3
10.0
1.79E-04
8.4
9.5
0.73

6.822E-02
1.568E+01
1.621E+00
2.249

SN =
MR (psi) =

2.71
27766

7.291E-02
1.468E+01
6.335E+00
2.104

SN =
MR (psi) =

1.73
81477

9.411E-02
1.137E+01
5.705E-01
1.630
1.534E-01

PSI

MR (kgf/cm2)
Average CBR
79131
3086
100
1491
30
500
5
CBRp subgrade =
3.8
HT (cm) =
78
SN =
4.415
RH =
0
MR (psi) =
7112.38

4.1
3.9
3.7
3.5
3.3
3.1
2.9
2.7

Model
HDM-III

9 10 11 12 13

Year

Figure 3 Results for the analysis of Case 2 of PSI prediction with time

Such requirement was not made when the AASHTO design formula was developed, as
the formula was derived with a single view of the pavement structure: MR as the subgrade
resilient modulus and SN associated with the whole pavement structure, comprised of the

20

surfacing, base and subbase layers. This requirement must be applied, however, for pavement
design using the AASHTO Guide, as the pavement layers must be determined imposing the
attendance of the design formula for every possible (SN, MR) combination. Therefore, the
AASHTO Guide forces the user to apply the design formula in ways that are only partially
supported by the experimental evidence at the AASHO Road Test, since the extension of the
original formula for other subgrade conditions was made through an analysis involving some
pavement sections with large granular base thicknesses, by considering them to be supported
on a high quality granular subgrade. In the model here developed, such procedure is
generalized by imposing that once the pavement performance is known, it must be predicted
by the model irrespective of the way the pavement structure is described by the pair (SN, MR).
This will lead to a model that can be rigorously applied for determination of all pavement
layers thicknesses.
Following the same procedure and using the same reference pavement sections, Table
6 shows the resulting calibration factors for the case of SCI prediction along time. Table 7 and
Figure 6 illustrate for one of the cases the calculations that were done in order to arrive at
these results. For the 16 reference cases, fatigue life of the HMAC surfacing layer at TR =
20% was nearly equal to the condition defined by SCI = 2.0 (associated with IGG = 70).
Table 6 shows that a calibration factor equal to FC = 4.73 can be adopted for SCI
predictions when considering the whole pavement structure supported by the subgrade soil,
with a coefficient of variation equal to 22%, a value consistent with the component
exclusively associated with performance predictions on the AASHTO overall standard
deviation S0. For the subbase analysis, a value for FC around 3.72 can be adopted, which
becomes FC = 3.15 for the base analysis, all of them with acceptable standard errors.
The cases on Table 6 showed a strong trend towards the SCI parameter being
explained by t0, the number of years for the first fatigue cracks appearance on the AC
surfacing layer, as shown on Figure 5. This is a reasonable result, since the onset of fatigue
cracking controls the IGG increase with time. This gives another procedure for SCI prediction
for overlay design, if a reliable model for t0 can be applied in order to predict the first
reflective cracks appearance at the overlay surface. In practice, it will be better to use the
relationship of Figure 5 for overlay design, since it can be expected a strong dependence of
SCI on the reflection cracking process.

21

Table 6 Results for SCI prediction on new asphalt pavements

Nyear
(AASHTO)
5.0E+05
5.0E+05
1.0E+06
1.0E+06
1.5E+06
7.0E+05
5.0E+05
3.0E+05
1.5E+06
1.0E+06
1.0E+06
1.5E+06
3.0E+05
3.0E+05
3.0E+05
7.0E+05

Case
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16

Subgrade
CBR
10
5
12
10
10
5
15
15
5
10
13
7
7
5
10
12

Design
period (years)
11
13
6
7
5
10
9
11
8
7
6
7
16
17
13
8
Average =
Std.Dev. =
PP
S0 =

Fc
Subgrade
4.800
6.200
3.800
4.100
4.000
5.600
3.400
4.300
4.900
4.100
3.750
5.300
6.900
6.300
4.250
4.000
4.731
1.043
0.220

Fc
Subbase
4.403
4.639
3.474
3.467
3.382
4.364
2.498
3.160
3.972
3.467
3.195
4.589
5.032
3.197
3.090
3.657
3.724
0.699
0.188

Fc
Base
4.201
3.390
3.631
3.297
3.217
2.813
2.836
3.587
2.263
3.297
3.734
2.481
5.122
1.415
1.262
3.822
3.148
0.978
0.311

7
y = 3.195x-0.891
R = 0.9944

t0 (years)

6
5
4
3
2
0.4

0.5

0.6

0.7

0.8

0.9

1.0

1.1

1.2

1.3

Alpha x Nyear (10^6)

Figure 5 Relationship between SCI and surfacing layer fatigue life

22

Table 7 Analysis for Case 2 for SCI prediction

Layer
HMAC
Granular Base
Granular Subbase
Subgrade
PP (years) =
HTp (DNER) =
Nyear =
QI0 =
PSI0 QI =
Np (USACE) =
h1_mn. =
VDMc =

h1 (cm)
10
18
40

ai
0.44
0.14
0.11

MR (kgf/cm2)
Average CBR
79131
3086
100
1491
30
500
5
CBRp subgrade =
3.8
HT (cm) =
78
SN =
4.415
RH =
0
MR (psi) =
7112.38

13
78.1 cm
5.00E+05
17
3.942
1.95E+07
10 cm
913

(Subgrade)
AlphaA =
W18 =
Bheta =
Fc =
Alpha =
(Subbase)
AlphaA =
W18 =
Bheta =
Fc =
(Base)
AlphaA =
W18 =
Bheta =
Fc =

SNC =
TYcr2 =
D0 mm =
Dmax =
Vs (TR=20%) =
Epst_HMAC =
Vf (years, TAI) =
Vf (years) =
Vf / Vs =

5.023
6.03
0.491
53.3
10.0
1.79E-04
8.4
9.5
0.73

2.181E-01
2.186E+01
1.621E+00
4.639

SN =
MR (psi) =

2.71
27766

2.984E-01
1.598E+01
6.335E+00
3.390

SN =
MR (psi) =

1.73
81477

1.632E-01
2.922E+01
5.705E-01
6.200
1.012E+00

SCI

5.0
4.5
4.0
3.5
3.0

Model
HDM-III

2.5
2.0
0

10 11

Year

Figure 6 Results for the analysis of Case 2 on SCI prediction with time

In the case of a rehabilitated pavement, this same model can be applied, since the
configuration of an asphalt concrete overlaid pavement resembles the performance prediction

23

made for a new pavement considering now the pavement structure as consisting only of the
surfacing layer supported by the base layer. Therefore:

(a) The Structural Number will be given by: SN = a1HR, where a1 = 0.44 is the asphalt
concrete overlay structural coefficient and HR is its thickness (in inches);
(b) The roadbed resilient modulus will now be given by:

MR =

E eq + (10 CBRef )

E eq = 2 1 2

2
pR
DC
[28]

where Eeq is calculated as the semi-infinite elastic space equivalent modulus of the pavement
structure (before rehabilitation), and DC is the pavement deflection, measured under a circular
applied load of pressure p and radius R, as in the FWD. The Poissons ratio of this
equivalent system can be fixed at an average typical value for pavement materials (such as
0.35). CBRef is the effective in situ CBR value of the upper layers of the pavement. DCP test
data could be employed for its evaluation on thin surfacing pavements, with a default value
CBRef = 150 for thick asphalt concrete surfacing layers over high quality crushed stone bases.
The idea behind considerations (a) and (b) is that deterioration of a rehabilitated
pavement is essentially the generation of distresses inside the applied asphalt concrete
overlay, with the old pavement behaving only as an elastic platform which gives a support
described by its equivalent elastic modulus MR, calculated by equation [28]. Such
consideration will lead to an underestimate of pavement performance only if the old pavement
has a significant remaining life, since in this case it cannot be degenerated simply as an
elastic platform. Practical applications of this model in such cases can be made by adding
an effective overlay thickness to the actual applied overlay thickness HR, with this effective
thickness calculated using the model for new pavements in order to reproduce its remaining
service life.
For the final overlay design, it will be also necessary to analyze the overlaid pavement
structure using the model for new pavements, in order to verify if there is a combination SN MR which is more critical to pavement performance than the one above described on items (a)
and (b). The layer materials in situ elastic moduli from Falling Weight Deflectometer tests

24

will be required for these analyses. This requirement will be critical if the old pavement has
been showing poor performance, if it has been left with extensive cracking for too long time
or if it is expected the passage in the future of significantly higher loads than in the past.
These are all situations under which the existing pavement can not be regarded as a stable,
consolidated platform, over which the HMAC overlay will be safely supported.
It is unlikely that the value of FC for a new pavement can also be applicable to a
rehabilitated one, since an existing pavement has already suffered the conditioning effects due
to traffic loads, as well as long-term climatic and ageing effects. It can be anticipated that FC
for a new pavement must be greater than the value for an overlaid one, since the conditioning
effects that the traffic loads apply to all pavement layers lead to a pavement structure more
stable than a new one, which has been subjected only to the stress history applied by the
compaction equipment.
The procedure here adopted was to consider the overlay case with the same view of a
new pavement, which means FC = 1.824 for PSI prediction and SCI = (106/Nyear)
(t0/3.2063)-1.125, as indicated on Figure 5.
The expected overlay age at first cracks appearance (t0) will be estimated from: t0 = Ff
Lf, where Lf is the fatigue life calculated applying the recalibrated T.A.I. fatigue law
(equation [21]), with t given by equation [22] and Ff = 0.919 is the fatigue calibration factor
from HDM cases, as shown in Table 8. These analyses were conducted using the followin
expression to calculate pavement deflection after application of the AC overlay:

0 = 8.68636 10 3 K ef0.9555741l 0.4315352

[29]
with r2 = 0.9761, SE = 0.1175181 and SE/SY = 15.4%, which is a regression developed from
several cases generated using FLAPS program. This formula was developed using
Westergaards equations for calculating the vertical deflection of a slab on a Winkler
foundation, where the effective reaction modulus of the existing pavement can be calculated
from the measured deflection by: Kef = p/c, with p = 5.8 kgf/cm2 under the 41 kN load on
FWD tests. Therefore, 0 = Q/(Kefl2)f(l), where Q is the applied load and l4 = E0HR3/[12(1-

02)Kef] is the fourth power of the radius of relative stiffness.

A calibration of the model for PSI prediction in the case of overlaid pavements cannot
be done using HDM empirical models, since these models could not discriminate between the
new pavements from the rehabilitated ones, except in the case of TYcr2 estimation (equations

25

[13] and [14]). Therefore, empirical data concerning the performance of overlaid pavements
will be plotted against the predictions of the model here developed to analyze its implications.

Table 8 Calibration of the reflection cracking fatigue model from HDM

6

HR YE4 (10 )
Dc
-2
(cm) (ESALs) (10 mm)
5
0.1
100
5
0.1
100
5
0.1
100
5
0.1
100
5
0.1
100
5
0.3
80
5
0.3
80
5
0.3
80
5
0.3
80
5
0.3
80
5
0.7
60
5
0.7
60
5
0.7
60
5
0.7
60
5
0.7
60
5
1.5
40
5
1.5
40
5
1.5
40
5
1.5
40
5
1.5
40
7.5
0.1
100
7.5
0.1
100
7.5
0.1
100
7.5
0.1
100
7.5
0.1
100
7.5
0.3
80
7.5
0.3
80
7.5
0.3
80
7.5
0.3
80
7.5
0.3
80
7.5
0.7
60
7.5
0.7
60
7.5
0.7
60
7.5
0.7
60
7.5
0.7
60
7.5
1.5
40
7.5
1.5
40
7.5
1.5
40
7.5
1.5
40
7.5
1.5
40
10
0.1
100
10
0.1
100
10
0.1
100
10
0.1
100
10
0.1
100

f
h1
Age
TR
MR
TYcr2
Vf
Ff
E0
t
2
2
-4
(Hz)
(cm) (years) (%)
(kgf/cm ) (kgf/cm ) (years) (10 ) (years)
1527
4.59 2.757
6.0
0.771
5
4
2 68.3 93746
5
7
5 68.3 93746
1527
4.50 2.757
6.0
0.756
5
10
10 68.3 93746
1527
4.36 2.757
6.0
0.732
5
13
20 68.3 93746
1527
4.10 2.757
6.0
0.689
5
16
50 68.3 93746
1527
3.52 2.757
6.0
0.592
7.5
4
2 68.3 93746
1909
4.78 2.278
3.7
1.284
7.5
7
5 68.3 93746
1909
4.69 2.278
3.7
1.260
7.5
10
10 68.3 93746
1909
4.54 2.278
3.7
1.222
7.5
13
20 68.3 93746
1909
4.29 2.278
3.7
1.154
7.5
16
50 68.3 93746
1909
3.71 2.278
3.7
0.998
10
4
2 68.3 93746
2545
4.89 1.798
3.5
1.410
10
7
5 68.3 93746
2545
4.80 1.798
3.5
1.384
10
10
10 68.3 93746
2545
4.66 1.798
3.5
1.344
10
13
20 68.3 93746
2545
4.41 1.798
3.5
1.270
10
16
50 68.3 93746
2545
3.83 1.798
3.5
1.103
12.5
4
2 68.3 93746
3817
5.01 1.276
5.0
1.002
12.5
7
5 68.3 93746
3817
4.92 1.276
5.0
0.984
12.5
10
10 68.3 93746
3817
4.78 1.276
5.0
0.956
12.5
13
20 68.3 93746
3817
4.53 1.276
5.0
0.905
12.5
16
50 68.3 93746
3817
3.95 1.276
5.0
0.789
5
4
2 56.7 85155
1527
5.97 2.500
8.9
0.670
5
7
5 56.7 85155
1527
5.84 2.500
8.9
0.655
5
10
10 56.7 85155
1527
5.63 2.500
8.9
0.632
5
13
20 56.7 85155
1527
5.25 2.500
8.9
0.589
5
16
50 56.7 85155
1527
4.39 2.500
8.9
0.493
7.5
4
2 56.7 85155
1909
6.15 2.073
5.5
1.118
7.5
7
5 56.7 85155
1909
6.02 2.073
5.5
1.094
7.5
10
10 56.7 85155
1909
5.81 2.073
5.5
1.056
7.5
13
20 56.7 85155
1909
5.44 2.073
5.5
0.987
7.5
16
50 56.7 85155
1909
4.58 2.073
5.5
0.831
10
4
2 56.7 85155
2545
6.27 1.650
5.0
1.253
10
7
5 56.7 85155
2545
6.14 1.650
5.0
1.227
10
10
10 56.7 85155
2545
5.93 1.650
5.0
1.185
10
13
20 56.7 85155
2545
5.55 1.650
5.0
1.110
10
16
50 56.7 85155
2545
4.69 1.650
5.0
0.938
12.5
4
2 56.7 85155
3817
6.39 1.192
6.8
0.939
12.5
7
5 56.7 85155
3817
6.25 1.192
6.8
0.920
12.5
10
10 56.7 85155
3817
6.04 1.192
6.8
0.889
12.5
13
20 56.7 85155
3817
5.67 1.192
6.8
0.834
12.5
16
50 56.7 85155
3817
4.81 1.192
6.8
0.707
5
4
2 48.5 79131
1527
7.85 2.307 12.4 0.635
5
7
5 48.5 79131
1527
7.65 2.307 12.4 0.619
5
10
10 48.5 79131
1527
7.34 2.307 12.4 0.594
5
13
20 48.5 79131
1527
6.78 2.307 12.4 0.548
5
16
50 48.5 79131
1527
5.51 2.307 12.4 0.446

26

Table 8 Calibration of the reflection cracking fatigue model from HDM (cont.)
6

HR YE4 (10 )
Dc
-2
(cm) (ESALs) (10 mm)
10
0.3
80
10
0.3
80
10
0.3
80
10
0.3
80
10
0.3
80
10
0.7
60
10
0.7
60
10
0.7
60
10
0.7
60
10
0.7
60
10
1.5
40
10
1.5
40
10
1.5
40
10
1.5
40
10
1.5
40
12.5
0.1
100
12.5
0.1
100
12.5
0.1
100
12.5
0.1
100
12.5
0.1
100
12.5
0.3
80
12.5
0.3
80
12.5
0.3
80
12.5
0.3
80
12.5
0.3
80
12.5
0.7
60
12.5
0.7
60
12.5
0.7
60
12.5
0.7
60
12.5
0.7
60
12.5
1.5
40
12.5
1.5
40
12.5
1.5
40
12.5
1.5
40
12.5
1.5
40

f
h1
Age
TR
(cm) (years) (%) (Hz)
7.5
4
2 48.5
7.5
7
5 48.5
7.5
10
10 48.5
7.5
13
20 48.5
7.5
16
50 48.5
10
4
2 48.5
10
7
5 48.5
10
10
10 48.5
10
13
20 48.5
10
16
50 48.5
12.5
4
2 48.5
12.5
7
5 48.5
12.5
10
10 48.5
12.5
13
20 48.5
12.5
16
50 48.5
5
4
2 42.3
5
7
5 42.3
5
10
10 42.3
5
13
20 42.3
5
16
50 42.3
7.5
4
2 42.3
7.5
7
5 42.3
7.5
10
10 42.3
7.5
13
20 42.3
7.5
16
50 42.3
10
4
2 42.3
10
7
5 42.3
10
10
10 42.3
10
13
20 42.3
10
16
50 42.3
12.5
4
2 42.3
12.5
7
5 42.3
12.5
10
10 42.3
12.5
13
20 42.3
12.5
16
50 42.3

MR
TYcr2
Vf
Ff
E0
t
2
2
-4
(kgf/cm ) (kgf/cm ) (years) (10 ) (years)
79131
1909
8.02 1.917
7.6
1.059
79131
1909
7.83 1.917
7.6
1.033
79131
1909
7.52 1.917
7.6
0.992
79131
1909
6.96 1.917
7.6
0.919
79131
1909
5.69 1.917
7.6
0.751
79131
2545
8.13 1.536
6.7
1.207
79131
2545
7.94 1.536
6.7
1.178
79131
2545
7.63 1.536
6.7
1.132
79131
2545
7.07 1.536
6.7
1.049
79131
2545
5.80 1.536
6.7
0.861
79131
3817
8.25 1.125
8.7
0.943
79131
3817
8.05 1.125
8.7
0.921
79131
3817
7.74 1.125
8.7
0.885
79131
3817
7.18 1.125
8.7
0.822
79131
3817
5.91 1.125
8.7
0.676
74622
1527
10.45 2.154 16.3 0.642
74622
1527
10.16 2.154 16.3 0.624
74622
1527
9.70 2.154 16.3 0.596
74622
1527
8.87 2.154 16.3 0.545
74622
1527
6.99 2.154 16.3 0.429
74622
1909
10.62 1.794
9.9
1.072
74622
1909
10.33 1.794
9.9
1.043
74622
1909
9.87 1.794
9.9
0.996
74622
1909
9.04 1.794
9.9
0.913
74622
1909
7.16 1.794
9.9
0.723
74622
2545
10.72 1.446
8.6
1.241
74622
2545
10.43 1.446
8.6
1.207
74622
2545
9.97 1.446
8.6
1.154
74622
2545
9.15 1.446
8.6
1.058
74622
2545
7.27 1.446
8.6
0.841
74622
3817
10.83 1.072 10.8 1.006
74622
3817
10.54 1.072 10.8 0.978
74622
3817
10.08 1.072 10.8 0.936
74622
3817
9.25 1.072 10.8 0.859
74622
3817
7.37 1.072 10.8 0.685
Average =
0.919
Std. Dev. =
0.239
Cv (%) =
26.0

One of such empirical models is the one by George et al. (1989), developed for
Mississippi highways:
VS = 64,43 AGE 0, 76 H R0,37

[30]

where: VS = rehabilitated pavement service life (years);

AGE = old pavement age at the time of rehabilitation, in years;
HR = HMAC overlay thickness (in).
27

This is an extremely simplistic model, with a high degree of masking of key

parameters influences. For instance, the decrease of pavement service life with the increase of
the old pavement age can be due to a higher level of pavement deterioration, which is not
described in the regression. Besides, there is not a parameter to describe the traffic rate.
Another model is the one from Hajek et al. (1987), also described in George et al.
(1989), which was developed for highway pavements in Ontario:
VS = 1,32 AGE 0,33 H R0, 47 DESAL0, 097 1,14 patch

[31]
where the pavement service life VS refers to a terminal PCR = 55, HR in mm, DESAL is the
daily number of ESALs and patch is equal to 0 under restricted routine maintenance
interventions and is equal to 1 under a regular practice. The traffic effect is not adequately
described in this model, since a 100% increase in it imply in a service life loss of only 10%.
Table 9 shows the results of these models to a set of selected design conditions. Pavement
service life for the model here developed was calculated for a terminal condition given by
SCIt = 3.0. The average values from the three models are near and around 14 to 16 years.
Figure 7 shows the comparison between the model for SCI prediction and the empirical
regressions here considered. The regression line is nearly identical with the equality line. It is
necessary to trace an envelope from 70.5% to 163% around the equality line to reach a
reliability of 70% for the predictions of overlaid pavement service life (60% for the
Mississippi data and 80% for the Ontario data). In spite of the shortcomings of the empirical
regressions [30] and [31], these results can be regarded as a support for the model here
developed.

30

y = 1.4069x0.9008
R = 0.821

Vs (years)

25
20
15
10
5

George et al.

Hajek et al.

0
0

10

15

20

25

30

35

Vs model (years)

Figure 7 Analysis of the model for SCI prediction

28

Table 9 Overlaid pavements analyzed for Vs at SCIt = 3.0

HR
(cm)
7
7
7
7
7
7
7
7
7
7
7
7
7
7
7
7
7
7
7
7
10
10
10
10
10
10
10
10
10
10
10
10
10
10
10
10
10
10
10
10
12.5
12.5
12.5
12.5
12.5
12.5
12.5
12.5
12.5

YE4 (10 )
(ESALs)
0.1
0.1
0.1
0.1
0.1
0.3
0.3
0.3
0.3
0.3
0.7
0.7
0.7
0.7
0.7
1.5
1.5
1.5
1.5
1.5
0.1
0.1
0.1
0.1
0.1
0.3
0.3
0.3
0.3
0.3
0.7
0.7
0.7
0.7
0.7
1.5
1.5
1.5
1.5
1.5
0.1
0.1
0.1
0.1
0.1
0.3
0.3
0.3
0.3

Dc
-2
(10 mm)
100
100
100
100
100
80
80
80
80
80
60
60
60
60
60
40
40
40
40
40
100
100
100
100
100
80
80
80
80
80
60
60
60
60
60
40
40
40
40
40
100
100
100
100
100
80
80
80
80

h1
(cm)
5
5
5
5
5
7.5
7.5
7.5
7.5
7.5
10
10
10
10
10
12.5
12.5
12.5
12.5
12.5
5
5
5
5
5
7.5
7.5
7.5
7.5
7.5
10
10
10
10
10
12.5
12.5
12.5
12.5
12.5
5
5
5
5
5
7.5
7.5
7.5
7.5

Age
(years)
7
10
15
20
25
7
10
15
20
25
7
10
15
20
25
7
10
15
20
25
7
10
15
20
25
7
10
15
20
25
7
10
15
20
25
7
10
15
20
25
7
10
15
20
25
7
10
15
20

TR
(%)
10
20
30
50
70
10
20
30
50
70
10
20
30
50
70
10
20
30
50
70
10
20
30
50
70
10
20
30
50
70
10
20
30
50
70
10
20
30
50
70
10
20
30
50
70
10
20
30
50

Vs model
(years)
11.9
11.9
11.9
11.9
11.9
6.9
6.9
6.9
6.9
6.9
6.2
6.2
6.2
6.2
6.2
8.9
8.9
8.9
8.9
8.9
18.6
18.6
18.6
18.6
18.6
10.7
10.7
10.7
10.7
10.7
9.4
9.4
9.4
9.4
9.4
12.6
12.6
12.6
12.6
12.6
25.3
25.3
25.3
25.3
25.3
14.5
14.5
14.5
14.5

Vs Eq.[25]
(years)
21.4
16.3
12.0
9.6
8.1
21.4
16.3
12.0
9.6
8.1
21.4
16.3
12.0
9.6
8.1
21.4
16.3
12.0
9.6
8.1
24.4
18.6
13.7
11.0
9.3
24.4
18.6
13.7
11.0
9.3
24.4
18.6
13.7
11.0
9.3
24.4
18.6
13.7
11.0
9.3
26.5
20.2
14.8
11.9
10.1
26.5
20.2
14.8
11.9

Vs Eq.[26]
(years)
10.7
12.1
13.8
15.2
16.3
9.6
10.8
12.4
13.6
14.7
8.9
10.0
11.4
12.6
13.5
8.2
9.3
10.6
11.7
12.5
12.7
14.3
16.3
17.9
19.3
11.4
12.8
14.7
16.1
17.3
10.5
11.8
13.5
14.8
16.0
9.7
11.0
12.5
13.8
14.8
14.1
15.8
18.1
19.9
21.4
12.7
14.2
16.3
17.9

29

Table 9 Overlaid pavements analyzed for Vs at SCIt = 3.0 (cont.)

HR
(cm)
12.5
12.5
12.5
12.5
12.5
12.5
12.5
12.5
12.5
12.5
12.5
14
14
14
14
14
14
14
14
14
14
14
14
14
14
14
14
14
14
14
14

YE4 (10 )
(ESALs)
0.3
0.7
0.7
0.7
0.7
0.7
1.5
1.5
1.5
1.5
1.5
0.1
0.1
0.1
0.1
0.1
0.3
0.3
0.3
0.3
0.3
0.7
0.7
0.7
0.7
0.7
1.5
1.5
1.5
1.5
1.5

Dc
-2
(10 mm)
80
60
60
60
60
60
40
40
40
40
40
100
100
100
100
100
80
80
80
80
80
60
60
60
60
60
40
40
40
40
40

h1
(cm)
7.5
10
10
10
10
10
12.5
12.5
12.5
12.5
12.5
5
5
5
5
5
7.5
7.5
7.5
7.5
7.5
10
10
10
10
10
12.5
12.5
12.5
12.5
12.5

Age
(years)
25
7
10
15
20
25
7
10
15
20
25
7
10
15
20
25
7
10
15
20
25
7
10
15
20
25
7
10
15
20
25

TR
(%)
70
10
20
30
50
70
10
20
30
50
70
10
20
30
50
70
10
20
30
50
70
10
20
30
50
70
10
20
30
50
70
Average =

Vs model
(years)
14.5
12.4
12.4
12.4
12.4
12.4
15.9
15.9
15.9
15.9
15.9
29.8
29.8
29.8
29.8
29.8
17.0
17.0
17.0
17.0
17.0
14.4
14.4
14.4
14.4
14.4
18.0
18.0
18.0
18.0
18.0
14.5

Vs Eq.[25]
(years)
10.1
26.5
20.2
14.8
11.9
10.1
26.5
20.2
14.8
11.9
10.1
27.6
21.1
15.5
12.4
10.5
27.6
21.1
15.5
12.4
10.5
27.6
21.1
15.5
12.4
10.5
27.6
21.1
15.5
12.4
10.5
15.7

Vs Eq.[26]
(years)
19.3
11.7
13.1
15.0
16.5
17.7
10.8
12.2
13.9
15.3
16.5
14.8
16.7
19.1
21.0
22.6
13.3
15.0
17.2
18.9
20.3
12.3
13.8
15.8
17.4
18.7
11.4
12.8
14.7
16.1
17.4
14.5

4 Case Studies
The model here developed was also applied to a set of independent field data in order to see if
it is capable of giving a reliable performance prediction. For a new pavement, highway SP070 (Figure 8) in Brazil was analyzed. The pavement was not designed solely using the
DNER pavement design method, but use was made also of mechanistic-empirical analyses.
The new highway SP-070 (segment Taubat Jacare in So Paulo state) was opened to
traffic in 1995 between km 61.3 and km 130.4. The average condition of the pavements in
2012 on lane 2, where commercial traffic was concentrated, was given by QI = 27 counts/km
and IGG = 9 to 14, under a traffic rate of 5 105 ESALs (AASHTO equivalency factors) per

30

year. Pavement deflection was at an average value of 27 10-2 mm under 40 kN (FWD), SN

= 3.6 and subgrade elastic modulus was MR = 217 MPa. Pavement structure consisted of 12
cm AC surfacing layer, 15 cm granular base layer (crushed stone) and 17 cm cemented-treated
crushed stone subbase, above a subgrade soil with CBR 15.

Figure 8 Analysis of the new pavement on SP-070 highway

In the case of rehabilitated pavements, an interesting outcome was the overlay of SP160 using 3 cm of a HMAC with asphalt-rubber. Figure 9 shows the performance prediction
for lanes 3 and 4 (where commercial traffic was concentrated). The pavement was overlaid in
2006. Longitudinal roughness was at QI = 13 counts/km in 2013 and reached QI = 19
counts/km, with IGG = 8 by the next year. It can be seen that this performance was essentially
predicted with success by the model. The fatigue model here calibrated predicted the first
appearance at the pavement surface of the cracks after 8 years, which is in agreement with the
observed pavement performance. The model was also in agreement with asphalt-rubber
equivalency factors from laboratory fatigue tests, which indicate 0.7 cm of HMAC with
asphalt-rubber for 1 cm of conventional HMAC mixes.

31

Figure 9 Overlaid pavement on SP-160 highway

A final verification of the model was performed applying it to pavement sections

designed using the TRRL Road Note 29 (Croney, 1977). Table 10 shows the derived
reliability levels and the pavement terminal conditions required for the pavement sections to
be in agreement with the model here developed. It is apparent that the design reliability level
increases and the pavement acceptable conditions during service lives around 16 to 22 years
become more stringent as the design traffic increases, in ways similar to the ones
recommended by the AASHTO Guide.
Figure 10 illustrates the procedure here adopted for the cases listed on Table 10.

Table 10 Application of the model to Road Note 29

Case
S6
S5
S2
S1
S4
S3

Np
(AASHTO)
1.00E+06
1.00E+06
1.00E+07
1.00E+07
1.00E+08
1.00E+08

Nyear
5.00E+04
5.00E+04
5.00E+05
5.00E+05
5.00E+06
5.00E+06

CBRsg
4
7
5
7
4
7

h1
(cm)
6.5
6.5
12
12
21
21

h2
(cm)
16
16
22
22
27
27

h3
(cm)
23
11
24
16
37
20

RL
(%)
50
50
70
75
85
90

QIt
(counts/km)
35
35
40
35
45
35

IGGt
80
80
40
30
30
30

Life
(years)
16
17
20
20
20
20

These results strongly support the validity of the proposed model, with its present
calibration factors, so that no reason exists up to this point to invalidate the model here
developed.

32

A final experiment was chosen for application of the model: the one conducted by
OCDE (1991) at Nantes Circular Test Track. On those accelerated tests, flexible and semirigid pavements were tested under a dual wheel, traveling at speeds ranging from 60.6 to 72.0
km/h. Of importance to the analysis here performed is the behavior of Structures I and II, both
of flexible pavements with a granular base layer (thickness equal to 28 cm) and asphalt
concrete thicknesses shown on Table 11, together with the number of load repetitions
recorded at the appearance of the first cracks at the pavement surface. These cracks were
transversal to the direction of traffic, indicating a mechanism of fatigue cracking produced by
tensile stresses at the bottom of the asphalt concrete surface layer. A dual axle load was
employed, with separation distance between wheels equal to 36 cm and wheel contact area
equal to 600 cm2. Passages-to-coverage ratios were greater than one due to the 1.60 m lateral
displacement of the dual wheel at a 45 minutes time interval.

Figure 10 Analysis of a Road Note 29 case

Table 11 also shows parameter a on: RD(N) = RD0 + a N0.5 for rut depth development
(mm) after N load repetitions. For Section I, RD = 25 mm after N = 2.5 105 load repetitions
at Q = 115 kN, with N = 4.9 105 at Q = 100 kN. For Section II, RD = 7 mm after N = 6.4
105 load repetitions at Q = 115 kN, and N = 3.6 105 at Q = 100 kN.

33

Table 11 Field data at Nantes Circular Test Track (OCDE, 1991)

Pavement Sector Axle load Speed
Section
No.
(kN)
(km/h)
I
01
115
72.0
I
11
100
60.6
II
02
115
72.0
II
12
100
60.6

Thickness
of A.C. layer (cm)
6.3
6.0
13.9
13.0

N0 (105)
First crack appearance
0.726
1.030
38.000
3.700

a
(103)
70.0
31.4
6.67
4.46

The tests were conducted with the asphalt concrete layer experiencing temperatures
between 8oC and 24oC. For determination of the pavement layers elastic moduli, pavement
deflections and tensile and compressive strains on the asphalt concrete layer were measured
under a moving axle load of 115 kN at 24oC. Values of the layers elastic moduli that were
found to be more consistent with these instrumentation parameters were:

For the asphalt concrete layer: E1 = 5610 MPa;

For the granular base:

E2 = 260 MPa on Section I and on Sector 02 of Section II
E2 = 190 MPa on Sector 12 of Section II

For the granular subgrade soil: E3 = 58 MPa.

Using these data, Figures 11 and 12 and Table 12 show the results from application of

the model here developed. Considering the first appearance on the surface of fatigue cracks as
associated with IGG = 20, good agreements were obtained for Section I, whereas for Section
II the predicted values are inside the experimental range, since 5.1 < (N0/N0I-01)model < 52.3.

Table 12 Performance prediction at Nantes Circular Test Track

Section
I
I
II
II

Sector
1
11
2
12

N0
7.260E+04
1.030E+05
3.800E+06
3.700E+05

N0_model
8.175E+04
1.278E+05
1.366E+06
1.491E+06
Average =

I-01

Rel = N0/N0
1.42
52.34
5.10
19.62

Rel_model
1.56
16.71
18.24
12.17

34

Figure 11 Performance prediction on Section I at Nantes Circular Test Track

35

Figure 12 Performance prediction on Section II at Nantes Circular Test Track

5 Application to Airfield Pavements

Extrapolating the AASHTO axle load equivalency factors to aircrafts, Figures 13, 14 and 15
depict the application of the model to three traffic conditions. The resulting pavement sections
are nearly the same as the ones designed using the FAA AC-150/5320-6E procedure for a 20
years design life, in spite of the caution required on such kind of analysis. This agreement
illustrates the power of the model and of AASHTO load equivalency factors. Also, application

36

of the model here developed shows that FAA minimum thickness requirement in the case of
light aircrafts is excessive.
For these cases, it was also calculated the Cumulative Damage Factor (CDF) using the
mechanistic interpretation of the USACE CBR formula, given by (Ahlvin,

Chou and

Hutchinson, 1974):
(1) Single Wheel:

max = 5,5 10 3 N 0,237

(2) Multiple Wheels:

max = 3,0 10 3 N 0,159

[32]
The results are presented on Table 13 and show that the adequate performance
predicted by the model is associated with CDF values well below 100%. This result gives
support to the model, as applied also to airfield pavements with AASHTO load equivalency
factors. This last result should not be a surprise if one considers the fact that AASHTO load
equivalency factors are in agreement with mechanistic ones for the case of fatigue cracking
analysis of asphalt concrete surfacing layers on flexible pavements.

Table 13 Airport pavements analyzed

Traffic
Light
Medium
Heavy

Aircraft
Q_wheel (kgf) L_wheel (cm) X (cm) Gamma_Subgr. Nadm
CDF (%)
ATR-72-200
5415.0
31.1
40.9
6.896E-04 1.037E+04
80.90
ATR-72-200
5415.0
31.1
40.9
2.474E-04 6.544E+06
0.32
EMB-145
11734.8
41.5
50.0
5.269E-04 5.635E+04
12.66
B 777-300
23727.4
39.6
89.8
5.213E-04 6.027E+04
28.67
A 320-100
15543.0
34.8
55.5
2.823E-04 2.854E+06
1.47

37

Performance prediction model

FAARFIELD
Figure 13 Application of the model to airfield pavements under light traffic conditions

38

Performance prediction model

FAARFIELD
Figure 14 Application of the model to airfield pavements (average traffic conditions)

39

Performance prediction model

FAARFIELD
Figure 15 Application of the model to airfield pavements (heavy traffic conditions)

References
AASHTO (1986). The AASHTO Guide for Design of Pavement Structures. American
Association of State Highway and Transportation Officials, Washington, DC.

40

Ahlvin, R.G., Chou, Y.T. & Hutchinson, R.L. 1974. Structural Analysis of Flexible Airfield
Pavements. Transportation Engineering Journal of ASCE, Vol. 100, No. TE3,
August, pp. 625-641.
Croney, D. and Croney P. The design and performance of road pavements. McGraw-Hill.
London. 1991. 2nd edn.
DNIT (1996). Manual de Pavimentao. Publicao IPR-719. Rio de Janeiro, Brasil.
George, K.P. and Rajagopal, A.S. (1989). Life Expectancy of Asphalt Overlays. Second
International Symposium on Pavement Evaluation and Overlay Design, Rio de
Janeiro. Vol. II, pp. 4.9.1 - 4.9.19.
Lytton, R.L. (1987). Concepts of pavement performance prediction and modeling. Second
North American Conference on Managing Pavements, Toronto, Canada. Vol. 2, pp.
2.1 2.19.
Paterson, W.D.O. (1987). Road Deterioration and Maintenance Effects: models for planning
and management. The World Bank, Washington, DC.
Pereira, A.M. (1979). "Consideraes sobre o procedimento de avaliao objetiva da
superfcie de pavimentos flexveis e semi-rgidos normalizado pelo DNER.
Fundamentos

metodolgicos.

Algumas

modificaes

possveis".

Simpsio

Internacional de Avaliao de Pavimentos e Projeto de Reforo, Anexos, Rio de

Janeiro, 9 a 14 de setembro, ABPv.
OCDE (1991). Essai OCDE en vraie grandeur des superstructures routires. Recherche en
matiere de routes et de transports routiers, DIRR No 125145, Paris, France.
Rodrigues, R.M. (2009). Modelos para previso de desempenho de pavimentos com base em
dados de concessionrias de rodovias. CBCR, Campinas.
The World Bank (1987). Description of the HDM-III Model. Volume I. Washington, DC.

41