Sei sulla pagina 1di 28

Christensen et al.

Hirsch Model for Estimating the Modulus of Asphalt Concrete



Donald W. Christensen
1
, Jr., Terhi Pellien
1
, and Ramon F. Bonaquist
2



Abstract


The purpose of this paper is to present a new, rational and effective
model for estimating the modulus of asphalt concrete using binder
modulus and volumetric composition. The model is based upon an
existing version of the law of mixtures, called the Hirsch model, which
combines series and parallel elements of phases. In applying the Hirsch
model to asphalt concrete, the relative proportion of material in parallel
arrangement, called the contact volume, is not constant but varies with
time and temperature. Several versions of the Hirsch model were
evaluated, included ones using mastic as the binder, and one in which the
effect of film thickness on asphalt binder modulus was incorporated into
the equation. The most effective model was the simplest, in which the
modulus of the asphalt concrete is directly estimated from binder
modulus, VMA, and VFA. Models are presented for both dynamic
complex shear modulus (|G*|), and dynamic complex extensional
modulus (|E*|). Semi-empirical equations are also presented for
estimating phase angle in shear loading and in extensional loading. The
proposed model was verified by comparing predicted modulus and phase
angles to values reported in the literature for a range of mixtures.

Key Words: Superpave, asphalt concrete, dynamic modulus, shear
modulus, models, law of mixtures.


Introduction


The purpose of this paper is to describe in detail the development and
use of a new model for predicting the modulus of asphalt concrete, which
is based upon an existing model for composite behavior called the Hirsch
model. This model was developed as part of NCHRP Projects 9-25 and
9-31, the objectives of which are to evaluate existing Superpave
requirements for volumetric composition and modify them if warranted.
The Hirsch model for hot mix asphalt concrete (HMAC) was developed
during the initial phases of these NCHRP projects to serve as a tool for
analyzing the effect of changes in air voids, voids in mineral aggregate
(VMA) and other volumetric mix factors on the modulus of asphalt


2

concrete and related mechanical properties. Although there are a number
of models for estimating the modulus of HMAC mixtures, these models
are for the most part empirical, they rely on obsolete semi-empirical
asphalt binder rheological data, and do not adequately utilize volumetric
composition in their formulation.
This paper presents the background needed to understand the
conceptual development of the modified Hirsch model, and how this
model has evolved to its current form. Several alternate formulations of
the model are presented: one that potentially accounts for the stiffening
effect of mineral filler within the mastic of the mixture; one that includes
a term for the effect of film thickness; and a simple model based directly
on volumetric properties. These three models for predicting asphalt
concrete modulus were evaluated using a substantial set of data from the
Federal Highway Administrations (FHWAs) Accelerated Loading
Facility (ALF) project, the MNRoad Project, and the WesTrack Project
(Pellinen, 2001). These data were fit to the models using non-linear least
squares. All three versions exhibited nearly identical accuracy, but
model parameters relating to both mastic modulus and film thickness
effects were not significant. Therefore, the simplest form of the Hirsch
model is recommended for general use by pavement engineers and
researchers in estimating modulus values of asphalt concrete mixtures.
This paper presents two versions of the Hirsch model for asphalt
concreteone for estimating the dynamic complex shear modulus |G*|,
and one for estimating the dynamic complex modulus in extension, |E*|.
Semi-empirical models are presented for estimating phase angles for
both loading methods based upon the Hirsch model. This paper includes
fairly extensive verification of the Hirsch model, based upon a number of
comparisons of predicted moduli values with measured values as
published in the literature. This paper is based upon appendices to
NCHRP Project 9-25 and Project 9-31 Interim Reports, but has been
edited for publication as a research paper (Christensen, 2001;
Christensen, 2002).


Background


The proposed method for estimating HMAC modulus is based upon a
widely used model for predicting properties of composite materialsthe
law of mixtures. As generally formulated, the law of mixtures represents
the mechanical response of two separate phases in parallel (Nichols,
1976):

E
c
= v
1
E
1
+ v
2
E
2
(1)


3


Where E refers to the modulus, or some other material property, v refers
to the volume fraction of a given phase, the subscript c refers to the
composite, and the subscripts 1 and 2 refer to different phases present in
the composite. A less common form of the law of mixtures can be
derived for phases in series (Nichols, 1976):

1/E
c
= v
1
/E
1
+ v
2
/E
2
(2)

Where the variables are as defined above. In the early 1960s, T. J.
Hirsch developed a variation of the law of mixtures for modeling the
mechanical behavior of asphalt concrete (Hirsch, 1962). The Hirsch
model combines parallel and series arrangement of the phases (Nichols,
1976):

1/E
c
= v
1s
/E
1
+ v
2s
/E
2
+ (v
1p
+v
2p
)
2
/(v
1p
E
1
+v
2p
E
2
) (3)

Where v
1s
and v
2s
refer to the volume fractions of phases 1 and 2,
respectively, in series arrangement, and v
1p
and v
2p
refer to the volume
fractions of phases 1 and 2, respectively, in parallel arrangement. An
equivalent expression for the Hirsch model can be formulated if it is
assumed that the relative proportions of phase 1 and phase 2 are the same
in the series and parallel portion of the model:
( )

,
_

+
+

,
_

+
2 2 1 1 2
2
1
1
1
1
1
E v E v
x
E
v
E
v
x
E
c
(4)
Where x is the ratio of phases in parallel arrangement to the total volume.
When x = 1, the Hirsch model produces results identical to completely
parallel phases (Equation 1), whereas when x = 0, it represents a pure
series arrangement (Equation 2). This model is therefore quite flexible,
and can be used to represent a wide range of composite behavior. Hirsch
found that for several portland cement concrete mixtures, x had a value
of about 0.5. These three composite models are shown in Figure 1.



4


V
1
V
2

V
1
V
2

V
V
V V
1s
2s
1p 2p

(a) Parallel phases (b) Series phases (c) Hirsch model

Figure 1. Schematic representation of composite models for parallel,
series, and Hirsch (combination) arrangement of phases.

Equation 3 is the basis for the proposed equation developed by the
research team for predicting asphalt concrete modulus from binder
modulus and volumetric properties. Asphalt concrete tends to behave
like a series composite at high temperature, but more like a parallel
composite at low temperature, and so the Hirsch model should be
appropriate for estimating the modulus of asphalt concrete. However, for
this model to be useful in modeling the modulus of asphalt concrete, the
relative proportions of the series and parallel phases must be time and
temperature dependent. The aggregate phase in the parallel portion of
the model is important in characterizing the behavior and performance of
HMAC, as it represents that portion of the aggregate particles in intimate
contact with each other; this portion of the aggregate is therefore called
the aggregate contact volume, Pc. In general, as the aggregate contact
volume increases, so will the modulus, strength, and resistance to
permanent deformation. High values of Pc indicate a very effective
structure producing good strengths and stiffness, typical at low
temperature. Low values of Pc tend to occur at high temperatures, and
indicate mixtures with low strength and stiffness.
Various modifications of the Hirsch model were devised and
evaluated in developing the final version. Four of the most important are
shown in Figure 2. In many of the early formulations of the model, it
was assumed that the binder in asphalt concrete was the mastic (mineral
filler plus binder) rather than binder alone. In Figure 2(a), the bulk of the
aggregate is in a series arrangement with a parallel combination of the
aggregate contact volume and mastic. Better results were obtained with
the arrangement shown in 2(b), where the aggregate contact volume is
combined in parallel with a series arrangement of the bulk aggregate and
mastic. In this figure, Vc refers to the aggregate contact volume, Va
refers to the aggregate volume exclusive of the contact volume, Vm is the


5

mastic volume, and Vv is the air void volume. For the configuration
shown in 2(a), the appropriate equation for dynamic modulus would be
as follows:

( )
1
2
' 1 '

1
]
1

+
VmEm PcEa
Va
Ea
Va
Ec (5)
Where:
Ec = composite (asphalt concrete) modulus
Va = volume fraction of aggregate, excluding contact volume
and mineral filler
Ea = aggregate modulus
Pc = aggregate contact volume, as volume fraction
Vm = volume fraction of mastic
Em = mastic modulus



Va
Vm Vc Vv

Va
Vm
Vc
Vv

(a) Series Version (b) Parallel Version

Vc
Vv Va,
Vm
Vap Vvp Vmp
Vas
Vms
Vvs

(c) Dispersed Version (d) Alternate Version

Figure 2. Schematic representation of four alternate versions of
modified Hirsch model.




6


For the configuration shown in Figure 2(b), the equation for the
composite modulus is given by the following equation:
( )
1
2
'
1

,
_

+ +
Em
Vm
Ea
Va
Vc VcEa Ec (6)
Where the variables are as defined previously. The arrangement in
2(b) can in fact be generalized by using an exponent m rather than
1 in the series portion of the model:
( ) ( )
m
m m m
VmEm Ea Va Vc VcEa Ec
1
1 1
' 1 + +


(7)
The exponent m can take any value from 1 to 1, though in this case
meaningful models would result only for values between 1 and about
0.2. For the case of perfect spheres of aggregates within the mastic
matrix, m = -0.5; this arrangement is shown in Figure 2(c). Thus,
Equation 7 represents a very flexible form of the Hirsch model.
In order to use the models represented in Figures 2(a), (b) and (c), the
modulus of the mastic must be estimated. The NCHRP 9-25 research
team has evaluated two approaches to calculating the modulus of the
mastic based upon the binder modulus and the volume fractions of binder
and mineral filler. The first is a version of the Einstein equation
developed by Shashidhar and his associates (Shashidhar et al., 1999):


( )
Eb
Vf CVf
AVf
Em
1
]
1

+
+

' ' 1 1
' 1
(8)
Where:
A = K
E
-1
K
E
= generalized Einstein coefficient
Vf = volume fraction filler in mastic = Vf/Vm
C = (1-Vf
max
)/Vf
max
2

Vf
max
= maximum volume fraction of filler in mastic

The Einstein equation is rational in form and appears to be
reasonably accurate. However, the predictions are quite sensitive
to the maximum volume fraction of filler (Vf
max
). As the volume
fraction of filler in the mastic approaches this maximum, the
predicted modulus of the mastic becomes extremely high; when
Vf = Vf
max
, the predicted value of Em is infinite. In reality, the
binder would most likely never actually incorporate this amount of
filler, as this represents a hypothetical maximum, at which the air
void content of the mastic would be exactly zero. Instead, a certain


7

amount of excess filler would occur throughout the mix as free
filler. However, this is difficult to account for within the
framework of a relatively simple method for predicting mixture
modulus.
An alternative approach involves the use of the generalized law of
mixtures:
( )
n
n n
Eb Vb Ea Vf Em
1
' ' + (9)

Where:
Vb = volume fraction binder in mastic = Vb/Vm
Eb = modulus of binder
n = exponent with values from 1 to +1

For dispersed systems, n will range from slightly less than 0 to about
0.5. A preliminary evaluation based upon typical binder-filler systems
has indicated that the predictions of Equation 9 are of the same
magnitude as those of Equation 8 when n = -0.2. Equation 9, because of
its relative simplicity and robustness, is attractive for use in the various
models. Results using this equation compared favorably with those
produced using the Einstein equation. Therefore much of the analyses of
the various forms of the Hirsch model therefore used Equation 9 for
estimating mastic modulus.
Preliminary analyses showed that these three versions of the Hirsch
model did not exhibit good accuracy. The version of the model that was
found to exhibit consistently good accuracy is shown in Figure 2(d).
This alternate formulation of the Hirsch modelis very similar to the
original model. The only difference in this version is that the series and
parallel sub-units of the model are combined in parallel, rather than in
series. This in effect places more emphasis on the parallel sub-unit of the
model. This version of the Hirsch model produced the best results, and
has the additional advantages of being relatively simple and very similar
to the original version of the model as formulated by Hirsch.
Mathematically, it can be expressed using the following equation:


( )
( )
1
2
2
'
'
'

1
]
1

+
+ + +
+ +
VmsEm
Vvs Vms
Ea
s Va
Vvs Vms s Va
VmpEm pEa Va Ec

(10)
Where the variables are as defined previously, but the addition of the
subscripts p and s denote parallel and series phases, respectively (see
figure 2(d)). As with the standard version of the Hirsch model, this


8

alternate formulation can be stated in somewhat simpler terms by using
the contact volume Pc to represent the proportion of parallel to total
phase volume:
( ) ( )
( )
1
2
'
1 '

1
]
1

+
+ + +
VmEm
Vv Vm
Ea
Va
Pc VmEm Ea Va Pc Ec (11)
Various types of functions where used to describe the contact factor
for use in Equation 11. Eventually, the research team found that the
following equation for the contact factor provided the best results:

1
1
'
'
2
0
P
P
VMA
Em VFM
P
VMA
Em VFM
P
Pc

,
_

,
_


+
(12)
Where VFM is the fraction of aggregate voids filled with mastic, VMA is
voids in the mineral aggregate, exclusive of mineral filler, and P
0
, P
1
,
and P
2
are empirically determined constants. Equation 12 is in fact a
type of logistic function, which produces a sigmoidal response in log-log
space typical of the behavior of many viscoelastic materials. The first
constant in Equation 12, P
0
, is directly related to the contact factor at
high temperatures and/or low frequencies; P
1
is an exponent related to
the rate of change of the contact factor with respect to the binder
modulus Em; P
2
is related to the location of the contact factor, which is
directly related to the overall stiffness of the asphalt concrete mixture.
As will be discussed later in this paper, preliminary analysis of the
mastic version of the Hirsch model demonstrated that it was accurate, but
also seemed to indicate that it was not necessary to consider the
stiffening effects of mastic. Instead, a simpler version of this model,
which treats asphalt concrete as a simple three-phase system of
aggregate, asphalt binder, and air voids, seemed appropriate:
( ) ( )
( )
1
2
1

1
]
1

+
+ + +
VbEb
Vv Vb
Ea
Va
Pc VbEb VaEa Pc Ec
(13)
Where Eb represents the binder modulus, and Vb represents the effective
binder volume. Note that the aggregate volumes in Equation 13 now
represent true aggregate volume including the volume of mineral filler.
The corresponding equation for the contact factor is identical to Equation
12, except that true VMA is substituted for VMA (voids plus binder
volume plus mineral filler volume).
A final modification of the Hirsch model was devised to determine if
film thickness could be incorporated into the expression for modulus.
The equation for this version of the Hirsch model is essentially identical


9

to 13, but the effective binder modulus, Eb, is substituted for the true
binder modulus:
( ) ( )
( )
1
2
'
1 '

1
]
1

+
+ + +
VbEb
Vv Vb
Ea
Va
Pc VbEb VaEa Pc Ec
(14)
The effective binder modulus is calculated by assuming that the
binder modulus at the aggregate surface is equivalent to the glassy
modulus, but then decreases to the normal binder modulus value over a
certain characteristic distance, t
T
, representing the transition zone
between the aggregate and binder. If the binder film thickness is called
t
F
, the equation for effective binder modulus is then given using the
following equation:

( ) Eb t E t t
EbE t
Eb
T g T F
g F
+
' (15)
Where E
g
is the glassy modulus of asphalt cement binder, assumed to be
1 GPa (145,000 lb/in
2
).
The reader should keep in mind that the above functions for the
Hirsch model, though stated in terms of the extensional modulus E, can
be just as easily formulated in terms of the shear modulus G.
Furthermore, the modulus values used in any of these equations can be
determined from creep, stress relaxation, or dynamic modulus tests. The
Hirsch model is a rational, though semi-empirical method of predicting
asphalt concrete modulusthat is, its structure is logical and based upon
the law of mixtures, but its use in practice requires calibration with
measured data. It should not be confused with a constitutive equation,
which rigorously delineates the relationships among stress, strain, and
material properties (such as modulus) in 2 or 3 dimensions. The Hirsch
model can however be used to estimate modulus values that are used in
various such constitutive equations.


Data used in Refining the Hirsch Model


In order to evaluate the various versions of the Hirsch model, and
refine the most promising of these models it was necessary to establish a
data base of modulus values for a wide range of mixtures. Such a data
set was created, based upon dynamic modulus measurements made at
Advanced Asphalt Technologies, LLC, (AAT) and at the Arizona State
University (ASU) as part of NCHRP Project 9-19 (Pellinen, 2001). The
AAT data set consists of dynamic shear modulus and phase angle data
collected using the SST, using the frequency sweep tests. The ASU data


10

set consists of dynamic compression modulus and phase angle data. The
mixtures tested were the same for both data sets, and originated from
FHWAs Accelerated Loading Facility (ALF) project, the MNRoad
Project, and the WesTrack Project; Pellinen has thoroughly summarized
these projects and the materials used in both data sets (Pellinen, 2001).
Inclusion of both shear and compression data allowed the research team
to develop versions of the Hirsch model for both cases, which is essential
since the relationship between |G*| and |E*| for asphalt concrete mixtures
is not at all straightforward nor well documented.
A summary of the database is given as Table 1. The database
includes results from testing on 18 mixtures using 8 different binders and
5 different aggregate sizes and gradations. A total of 206 observations
are included in the dataset for each type of measurement (shear and
compression). A wide range of volumetric compositions is also
represented, although one shortcoming in this data is the lack of mixtures
with low air voids and low VFA. The data set is however extensive, and
suitable for initial development of the Hirsch model. It is possible that
larger and more robust data sets could be used in the future to further
refine this method for estimating asphalt concrete modulus.


Method of Analysis


The general approach used in evaluating the various versions of the
Hirsch model, and refinement of the most promising was non-linear least
squares (Chapra and Canale, 1988). Christensen has presented a detailed
description of the use of this technique in analyzing IDT creep data
(Christensen, 1998). In general terms, this procedure uses an iterative,
numerical procedure to calculate the values for parameters in a function
so that the sum of the square of the error terms is minimized. Graphical
techniques were used to identify and eliminate several records that were
clearly outliers, and to determine if other potential predictor variables
existed which were not included in the model. All of the outliers
eliminated from the data were high-temperature measurements made
using the SST at very low stress levels. These measurements were
considered to be unreliable, since the very low stress levels used were
difficult to measure and appeared to be quite noisy.
After refinement of the final versions of the Hirsch model, simple
curve fitting was used to develop a function for phase angle. Such
equations are sometimes needed for calculation of storage and loss
modulus values.



11


Table 1. Summary of Database of Mixture Modulus Data

Factor ALF MNRoad WesTrack Total
Design Method Marshall Marshall Superpave N/A
Binders AC-5, 10, 20
SBS-modified
PE-modified
AC-20
120/150-
PEN
PG-64-22 8
Aggregate Sizes
and Gradations
19-mm dense
37.5-mm fine
9.5-mm
fine
19-mm fine
19-mm coarse
5
Mixtures 7 5 6 18
Total Data
Points
78 59 69 206
For Complete Database
Voids, Vol. % 5.6 to 11.2
VMA, Vol. % 13.7 to 21.6
VFA, % 38.7 to 68.0
Temperatures,
C
4, 21, 38
Frequencies 0.1 and 5
|G*|, MPa 20.0 to 3,880
|E*|, MPa 183 to 20,900
Phase angle,
degrees
8 to 61



12


Results


The three most promising versions of the Hirsch model were all
variations of the alternate formulation shown graphically in Figure 2(d),
and represented in its basic form in Equation 10. The three versions of
this model are called here the mastic version (Equation 11), the simple
version (Equation 13), and the transition zone version (Equations 14 and
15). In all cases, Equation 12 was used to characterize the contact factor
Pc.

Model for |G*|

The results of the non-linear least squares analyses of these three
models for dynamic complex shear modulus (|G*| as measured using the
SST) are summarized in Table 2. The value for parameter P
0
could not
be determined reliably using the least-squares procedure, probably
because the database did not include enough values at high temperatures
and low frequencies. However, it is important to include at least an
estimate for this parameter, since it represents the limiting asphalt
concrete modulus at high temperatures and/or low frequencies.
Comparing model predictions with published master curves for asphalt
concrete, it was estimated that an appropriate value for this constant is
about 3.
All three versions of the Hirsch model exhibited identical r
2
values of
96.8 %, suggesting that the more complex versions of the model,
intended to account for mineral-filler effects and the effect of film
thickness, are no more effective than the model treating asphalt concrete
as a simple three-phase system. For the mastic model, the law of
mixtures exponent for the mastic, n (Equation 9) has an unrealistic value
of 6.9, and a standard error of 281 %, indicating that this parameter
does not contribute significantly to the accuracy of the model. For the
transition zone model, similar problems are observed in the estimate of
the transition zone thickness; t
T
is estimated at only 0.03 microns, and
has a huge standard error of estimate of almost 4,000 %. It is therefore
concluded that the most effective version of the Hirsch model is the
relative simple version represented by Equation 13, which can be given
in terms of the complex shear modulus, |G*|, VFA, and VMA:


13



Table 2. Summary of Least-Squares Estimation of
Parameters for |G*| Hirs ch Model.

Mastic Model
Equation(s): 11 & 12
Parameter Estimate Std. Error
Ea, lb/in
2
635,000 7.2 %
P
0
3 N/A
P
1
0.678 2.4 %
P
2
396 10.4 %
n (Eqn. 9) 6.9 281%
t
T,
microns N/A N/A
r
2
96.8 % N/A

Simple Model
Equation(s): 13 & 12
Parameter Estimate Std. Error
Ea, lb/in
2
601,000 7.1 %
P
0
3 N/A
P
1
0.678 2.4 %
P
2
396 6.9 %
n (Eqn. 9) N/A N/A
t
T,
microns N/A N/A
r
2
96.8 % N/A

Transition Zone Model
Equation(s): 14, 15 & 12
Parameter Estimate Std. Error
Ea, lb/in
2
601,000 7.2 %
P
0
3 N/A
P
1
0.678 2.4 %
P
2
452 8.7 %
n (Eqn. 9) N/A N/A
t
T,
microns 0.03 3,760 %
r
2
96.8 % N/A


14

( )
( )
1
* 000 , 601
100 1
1
000 , 10
* 100 1 000 , 601 *

1
1
]
1

+
1
]
1

,
_


+
binder
binder mix
G VFA
VMA VMA
Pc
VMA VFA
G VMA Pc G

(16)
Where |G*|
mix
is the complex shear modulus for the mixture, and
|G*|
binder
is the complex shear modulus for the binder, both in units of
lb/in
2
, and VFA and VMA are both given as percentages. The binder
modulus can either be determined experimentally using the dynamic
shear rheometer (DSR) or similar device, or can be estimated from one
of several available mathematical models. It should be at the same
temperature and loading time selected for the mixture modulus, and in
consistent units. The contact factor is given by the following function:

678 . 0
678 . 0
*
396
*
3

,
_

,
_

VMA
G VFA
VMA
G VFA
Pc
binder
binder
(17)
Figure 3 shows the |G*| values predicted using Equations 16 and 17
versus measured values. There is generally good agreement.
1000
10000
100000
1000000
1000 10000 100000 1000000
Measured |G*|, psi
P
r
e
d
i
c
t
e
d

|
G
*
|
,

p
s
i

Figure 3. Predicted Versus Measured Shear Complex Modulus (r
2
=
96.8 %); Solid Line Represents Equality.


15

Model for |E*|

Non-linear least squares analysis was also performed using the |E*|
data gathered by Pellinen at Arizona State University as part of NCHRP
Project 9-19 (Pellinen, 2001). In the initial calibration of the model for
|E*|, performed during Phase I of NCHRP Project 9-25, the data set used
coincided with that developed for the |G*| model. However, subsequent
evaluation of the model showed that results at extreme high and low
temperatures were not always accurate. Therefore, as part of Phase I of
NCHRP Project 9-31, an expanded data set was created, which included
additional data at 9 and 54 C. This resulted in a somewhat greater
range of values for both |E*| and phase angle in the data set.
Furthermore, in the expanded data set replicate measurements were
averaged, partly because of the greatly expanded size of the resulting
data set, but also to minimize variability due to experimental error. It
was felt that this approach would provide a better estimate of the
accuracy of the model. The final values for the Hirsch model parameters
for dynamic complex modulus in extension were estimated to be as
follows:

Ea: 4,230,000 psi (t 6.5 %)
P
0
: 649 (t 9.0 %)
P
1
: 19.7 (t 20.7 %)
P
2
: .575 (t 3.0 %)

The r
2
value for the |E*| data (98.2 %) is slightly higher than for shear
data. Also, note the much higher value for Ea, which is expected as
compression moduli for most materials, including asphalt concrete, are
almost always much higher than shear moduli. The other parameters are
similar to the values determined for the shear model. The predicted and
measured values for |E*| are shown in Figure 4; residuals as a function of
predicted |E*| are shown in Figure 5. Applying appropriate rounding to
the coefficients listed above, the pertinent equation for compression
modulus is as follows:

( )
( )
1
* 3 000 , 200 , 4
100 1
1
000 , 10
* 3 100 1 000 , 200 , 4 *

1
1
]
1

+
1
]
1

,
_


+
binder
binder mix
G VFA
VMA VMA
Pc
VMA VFA
G VMA Pc E

(18)


16

1.0E+04
1.0E+05
1.0E+06
1.0E+07
1.0E+04 1.0E+05 1.0E+06 1.0E+07
Measured |E*|, psi
P
r
e
d
i
c
t
e
d

|
E
*
|
,

p
s
i

Figure 4. Predicted and Measured values for Complex Modulus in
Compression (r
2
= 98.2 %); Solid Line Represents Equality.

-0.5
-0.4
-0.3
-0.2
-0.1
0.0
0.1
0.2
0.3
0.4
0.5
1.0E+04 1.0E+05 1.0E+06 1.0E+07
Predicted |E*|, psi
R
e
s
i
d
u
a
l

(
L
o
g

|
E
*
|
)

Figure 5. Residuals (Predicted-Measured) Values of Log |E*| for
Hirsch Model.

Note that the binder shear modulus in Equation 18 is multiplied
by 3 as an estimate of the extensional modulus: |E*|binder
3|G*|binder. This is based on an assumption of incompressibility,
that is, that Poissons ratio is 0.5. The following function is used
for estimating the contact volume for use in conjunction with
Equation 18:


17


58 . 0
58 . 0
* 3
650
* 3
20

,
_

,
_

VMA
G VFA
VMA
G VFA
Pc
binder
binder
(19)
Prediction of Phase Angle

As discussed previously, in addition to a need for estimating modulus
in shear and compression, there is also a need to estimate the phase angle
from compositional data. For example, the fatigue models developed
during the Strategic Highway Research Program (SHRP) used the loss
modulus as a predictor variable, which is a function of both the complex
modulus and the phase angle (Taybali et al., 1995). The research team
found a good empirical relationship between the log of the contact factor
Pc (Equation 12) and the phase angle. Plots showing these relationships,
which includes the empirically determined equations for phase angle as a
function of log (Pc), are shown in Figures 6 and 7. The following
function can be used to estimate phase angle using Pc as determined
from shear data (r
2
= 82.9 %):
( ) 6 . 9 log 39 log 5 . 9
2
+ Pc Pc (20)

y = -9.4633x
2
- 38.834x + 9.6031
R
2
= 0.8294
0
10
20
30
40
50
60
70
-3.0 -2.5 -2.0 -1.5 -1.0 -0.5 0.0
Log (Pc)
P
h
a
s
e

A
n
g
l
e
,

D
e
g
r
e
e
s

Figure 6. Phase Angle as a Function of Log (Pc), Shear Data.




18

y = -20.56x
2
- 54.619x
R
2
= 0.8906
0.00
10.00
20.00
30.00
40.00
50.00
-3.0 -2.5 -2.0 -1.5 -1.0 -0.5 0.0
Log (Pc)
P
h
a
s
e

A
n
g
l
e
,

D
e
g
r
e
e
s

Figure 7. Phase Angle as a Function of Log (Pc), Compression Data.


The corresponding function for estimating from compression data is
given by the following Equation (r
2
= 89 %):

( ) Pc Pc log 55 log 21
2
(21)

These relationships, though not highly accurate, are useful for
estimating phase angle from asphalt concrete volumetric composition. It
should be kept in mind that the measurement of phase angle is
particularly difficult, so the amount of scatter in Figures 6 and 7 is not
surprising. There are two significant differences in these plots. The
phase angle in shear (Figure 6) has a value of 9.6 when Pc is zero; this
indicates that in shear measurements, the phase angle is about 10 degrees
even at very low temperatures and high frequencies. For the extensional
data, this is not truethe phase angle is zero when Pc is zero. It is
possible that the non-zero phase angle when Pc is zero for shear data is
an artifact, caused by friction in the horizontal bearing used in the SST
system (there are no bearings in the compression tests, outside of those
inherent in the actuators). The other significant difference is that for the
compression data, there is a clear maximum in phase angle, whereas the
shear data shows no maximum. In general, it appears that the phase
angle data in extension is somewhat more reasonable than that
determined in shear.




19

Verification of the Model


In order to verify the final version of the Hirsch model in as
independent a manner as possible within the limited time available
during the initial phases of NCHRP Projects 9-25 and 9-31, the models
for shear modulus (Equations 16, 17 and 20) were used to estimate shear
modulus for data as reported by Alavi and Monismith in research related
to the original SHRP program (Alavi and Monismith, 1994). These
measurements were made using a cylindrical shear test, a completely
different technique than that upon which the model was developed.
Alavi and Monismith used one aggregate (19-mm nominal maximum
size) and one binder, but a range of asphalt and air void contents. Binder
modulus values were estimated using the model developed by
Christensen and Anderson for SHRP binders (Christensen and Anderson,
1992). Some volumetric information was estimated because of the
limited data reported by Alavi and Monismith. The predicted and
measured values of shear modulus are shown in Figure 8; the agreement
is good, though the Hirsch model appears to slightly under-predict
modulus values at higher levels. In Figure 9, predicted and measured
phase angles are plotted for these same data. The agreement here is not
as good, especially at higher phase angles. In Alavi and Monismiths
data, the phase angle value, after peaking, decreases to very low values.
In the SST shear data, on the other hand, the phase angle increases at a
decreasing rate to a maximum value without any subsequent decrease.
The research team believes this is an inherent difference in the
measurement methods. Phase angle predictions using the Hirsch model
are probably most accurate at low temperatures and/or high frequencies.
As a second verification of the model, the complex modulus in
compression (|E*|) was predicted using both the Hirsch model and using
Andrei and Witczaks equation, and compared to the measured values
reported by Alavi and Monismith (Andrei Witczak, 1999; Alavi and
Monismith, 1994). This comparison is shown graphically in Figure 10.
The values for the Hirsch model are in excellent agreement, whereas
Andrei and Witczaks equation slightly under-predicts at higher modulus
values. Although this comparison is too limited to make broad
generalizations, it suggests that the Hirsch model is in general agreement
with Andrei and Witczaks equation, and is at least as accurate. Hirsch
model predictions for |E*| are probably in better agreement with Alavi
and Monismiths data, compared to the |G*| predictions, because the
experimental technique used by Alavi and Monismith for the shear
measurements was substantially different than that used by AAT in
making the SST measurements. In contrast, Pellinens compression


20

moduli were probably determined using methods giving results
comparable to Alavi and Monismiths.

10
100
1000
10000
100000
10 100 1000 10000
Predicted |G*|, MPa
M
e
a
s
u
r
e
d

|
G
*
|
,

M
P
a

Figure 8. Measured Complex Shear Modulus and Values Predicted
Using the Hirsch Model; Solid Line Represents Equality (data from
Alavi and Monismith, 1994).
0
10
20
30
40
50
60
70
0 10 20 30 40 50 60 70
Predicted Phase Angle, Degrees
M
e
a
s
u
r
e
d

P
h
a
s
e

A
n
g
l
e
,

D
e
g
r
e
e
s

Figure 9. Measured Phase Angle s and Values Predicted Using the
Hirsch Model; Solid Line Represents Equality (data from Alavi and
Monismith, 1994).




21

100
1000
10000
100000
100 1000 10000 100000
Measured |E*|, MPa
P
r
e
d
i
c
t
e
d

|
E
*
|
,

M
P
a
Witczak Model
Hirsch Model
Equality

Figure 10. Predicted and Measured |E*| Values; Solid Line
Represents Equality (data from Alavi and Monismith, 1994).


Another comparison useful for verification of the Hirsch model is
shown in Figure 11, which is a master curve for mixture V0W1 from
Alavi and Monismiths study (Alavi and Monismith, 1994). This figure
shows predicted and measured values for |E*| and phase angle as a
function of reduced frequency at 40 C. This figure confirms that the
frequency dependence and general shape of the functions for complex
modulus and phase angle as predicted by the Hirsch model are
reasonable and in good agreement with experimental values. However, it
appears that the phase angle predictions do not vary quite as strongly
with frequency as the measured values.
A more thorough verification of the Hirsch model was performed
using data from a recent sensitivity study by Witczak and his associates
(Witczak et al., 2001). In this study, |E*| measurements were made on a
range of mix variations based upon an Arizona Department of
Transportation mixture. This asphalt concrete used a 25-mm nominal
maximum aggregate size blend with an PG 64-22 asphalt. The basic mix
design was varied using four different target air void levels (1.5, 4, 7 and
10 %) and four binder content levels (3.9, 4.55, 5.2, and 5.9 %). In
addition to measured |E*| values, modulus values were also predicted
using Witczaks equation. In Figure 12, |E*| values predicted using the
Hirsch model are compared to measured values reported by Witczak and
his team. The standard error in this case is 41 %, which is slightly better
than the standard error for Witczaks model (45 %), but about double


22

that for the experimental error for these data (20 %). Although not as
good as actual measurements, the accuracy of the model predictions is
probably suitable for many practical design and analysis applications.
Estimated modulus values in fact are probably similar in reliability to
measured values when only limited replicate tests can be performed, or
when laboratory personnel are not experienced in making modulus
measurements on asphalt concrete specimens.
100
1000
10000
100000
1.E-04 1.E-02 1.E+00 1.E+02 1.E+04 1.E+06
Reduced Frequency at 40 C, Hz
|
E
*
|
,

M
P
a
0
10
20
30
40
50
60
70
80
90
P
h
a
s
e

A
n
g
l
e
,

d
e
g
r
e
e
s
Meas. |E*|
Pred. |E*|
Meas. Phase
Pred. Phase

Figure 11. Master Curve for Mixture V0W1 as Reported by Alavi
and Monismith, and as Predicted Using the Hirsch Model.

A second comparison with these data is shown in Figure 13, which is
a plot of modulus values predicted using the Hirsch model against values
determined using Witczaks equation. In this case, the agreement is
extremely close. However, such close agreement between these models
will not always occur; in this case, it is probably fortuitous and brought
about by close agreement between the SHRP binder data used in the
Hirsch model and the routine data used in Witczaks equation. Given the
close agreement between these models, and the similar standard errors
compared to the measured |E*| values, this confirms the finding that the
Hirsch model and Witczaks equation are of similar accuracy and in
reasonably close agreement. As pointed out earlier, the Hirsch model is
simpler and more rational, and makes use of |G*| data as produced using
standard Superpave binder tests..



23

10
100
1000
10000
10 100 1000 10000
Measured |E*|, ksi
P
r
e
d
i
c
t
e
d

|
E
*
|
,

k
s
i

Figure 12. Predicted and Measured Dynamic Modulus Values Using
Data from NCHRP 9-19 Sensitivity Study; Solid Line Represents
Equality (Witczak et al., 2001).



10
100
1000
10000
10 100 1000 10000
Predicted |E*| (ksi), Witczak Model
P
r
e
d
i
c
t
e
d

|
E
*
|

(
k
s
i
)
,

H
i
r
s
c
h

M
o
d
e
l

Figure 13. Dynamic Modulus Values Predicted Using the Hirsch
Model and Using Witczaks Equation, NCHRP 9-19 Sensitivity
Data; Solid Line Represents Equality (Witczak et al., 2001).



24

A final verification of the Hirsch model was done using bending
beam data and mixture creep compliance data published as part of the
SHRP project (Lytton et al., 1993). Mixture creep compliance values
were determined on a range of field cores using the indirect tensile creep
tests. Using reported binder creep modulus values as determined using
the bending beam rheometer, and reported volumetric composition, the
Hirsch model was used to predict creep modulus values, which were
compared to values reported by Lytton and associates. The results of this
comparison are shown in Figure 14. The Hirsch model appears to under-
predict the modulus values at high stiffness levels, but at this point the
mixture creep modulus is very high, in the range of 2 to 6 million lb/in
2
,
and approaching the glassy limit. At values below about 2 million lb/in
2
,
the predicted values are in reasonably good agreement with the measured
values. Comparison of Figure 14 with Figure 12 show that these
comparisons are in agreement; the under-prediction seen with the Hirsch
model at very high modulus values is probably a function of the
estimated constant glassy modulus value (4,200,000 lb/in
2
). From a
practical perspective, once an asphalt concrete mixture is exhibiting
modulus values in this range, it will be extremely stiff and brittle, and
would be probably subject to high levels of thermal cracking and fatigue
cracking if used in a pavement in this condition. This discrepancy
should therefore not be considered a significant problem.

100
1000
10000
100 1000 10000
Measured Creep Modulus, ksi
P
r
e
d
i
c
t
e
d

C
r
e
e
p

M
o
d
u
l
u
s
,

k
s
i5 seconds
100 seconds
equality

Figure 14. Predicted Creep Modulus Values Compared with
Measured Values, Using SHRP A-005 Low-Temperature Data; Solid
Line Represents Equality (Lytton et al., 1993).



25

Comparison of |E*| and |G*| Values

A final important comparison can be made between the measured
moduli in shear and compression for the entire data set used to develop
the Hirsch model. Recall that the shear moduli were measured at AAT
using the SST frequency sweep procedure, while the compression moduli
were determined entirely independently at Arizona State University,
though the same materials were used. This allows a direct comparison of
independently determined modulus values. A plot of |G*| versus |E*| for
these data is shown in Figure 15. Two important observations can be
made concerning this figure. First, the R
2
value for this relationship,
92.9 %, is actually lower than the values for the predicted and measured
moduli for both shear and compression data. This indicates that there is
a large amount of noise in the modulus measurements. Because of the
relatively poor precision of modulus determinations on asphalt concrete,
it is possible that modulus predictions using the Hirsch model (or other
similarly accurate procedure) might be nearly as accurate as independent
modulus measurements made on the same mixture. Thus for many
practical applications, using modulus values predicted using an accurate
model is probably just as effective as using measured values, and of
course much less time consuming and expensive.

y = 0.0603x
1.0887
R
2
= 0.9291
1000
10000
100000
1000000
10000 100000 1000000 10000000
Measured |E*|, psi
M
e
a
s
u
r
e
d

|
G
*
|
,

p
s
i
|G*| = |E*|/2.8

Figure 15. Comparison of Measured Modulus in Shear (|G*|) and
Compression (|E*|), for ALF, MNRoad, and WesTrack Data.



26

A second important observation concerning Figure 15 is the
relationship between |G*| and |E*|. Paving engineers might be tempted
to try to convert compression moduli to shear moduli using the simple
relationship G = E/(1+), where is Poissons ratio. If a value of 0.4 is
assumed for , this suggests that |G*| |E*|/2.8. The line representing
this equation is included on Figure 15, and clearly shows that this
approximation is not at all accurate. Although the SST is certainly a far
from ideal test system, the agreement between SST-based predictions
and Alavi and Monismiths data (Figure 5) suggests that the relationship
shown in Figure 15 is for the most part real. The inaccuracy of simple
conversions between shear and compression moduli is probably due to
several factors, including non-linearity. However, non-linearity is
generally not significant at low temperatures and high frequencies, but as
seen in Figure 15, the simple conversion is not accurate even at high
modulus values. This discrepancy is probably caused in large part by
anisotropy in the mechanical behavior of asphalt concrete. Other
researchers have found similar discrepancies between HMAC modulus
values made using different loading geometries (Taybali et al., 1994)
Therefore, engineers should use caution in selecting modulus test data or
predicted modulus values for use in pavement design and analysis. For
example, Bonnaures fatigue equations were developed on the basis of
flexural modulus data; for best accuracy, the modulus values used in
conjunction with his fatigue equation should therefore be values based on
flexural measurements, or if these are not available, extensional data.
Engineers should not rely on conversions between shear, compression,
and flexural data based upon linear elastic theory and assumptions of
homogeneity and anisotropy. Although not as elegant, empirically
determined relationships such as that illustrated here and reported by
other researchers are likely to be more reliable.


Conclusions


A relatively simple version of the Hirsch model for composite
behavior has been developed for estimating the complex modulus
and phase angle of asphalt concrete in shear and compression.
The Hirsch model is both simpler and more rational than existing
models for predicting modulus, and requires as input only asphalt
concrete volumetric composition and SHRP binder data, and so is
ideal for use in examining the relationships among HMAC
volumetrics, modulus, and related aspects of pavement performance


27

The model appears to be in good agreement with Andrei and
Witczaks model, and is of similar accuracy.
Modulus values predicted using the Hirsch model are potentially
nearly as accurate as measured modulus values. For many pavement
design and analysis procedures, predicted modulus values can be
effectively used, and can be determined much more quickly and
cheaply. Using predicted modulus values should be considered
when reliable measurements by experienced lab personnel are not
available.
Relationship among modulus values determined using different test
methods are complex and cannot be accurately predicted using
simple linear elastic theory and assumptions of homogeneity and
anisotropy. Engineers should be careful to select the appropriate
modulus value for their intended purpose, whether that value is
measured or predicted using one of the available models.

References


Alavi, S. H., and C. L. Monismith, Time and Temperature Dependent
Properties of Asphalt Concrete Mixes Tested as Hollow Cylinders and
Subjected to Dynamic Axial and Shear Loads, Journal of the
Association of Asphalt Paving Technologists, Vol. 63, 1994, pp. 152-
175.

Andrei, D., Witczak, M.W., and Mirza, W., Development of a Revised
Predictive Model for the Dynamic (Complex) Modulus of Asphalt
Mixtures, NCHRP 1-37A Inter Team Technical Report, University of
Maryland, March 1999.

Bonnaure, F. P., A. H. J. J. Huibers, and A. Boonders, A Laboratory
Investigation of the Influence of Rest Periods on the Fatigue
Characteristics of Bituminous Mixes, Proceedings, the Association of
Asphalt Paving Technologists, Vol. 51, 1980, p. 104.

Bonnaure, F., G. Gest, G. Gravois, and P. Uge, A New Method of
Predicting the Stiffness of Asphalt Paving Mixtures, Proceedings, the
Association of Asphalt Paving Technologists, Vol. 46, 1977, p. 64.

Chapra, S. C., and R. P. Canale, Numerical Methods for Engineers, New
York: McGraw-Hill, Inc., 1988, 812 pp.



28

Christensen, D. W., Analysis of Creep Data from Indirect Tension Test
on Asphalt Concrete, Journal of the Association of Asphalt Paving
Technologists, Vol. 67, 1998, pp. 458-489.

Christensen, D. W., NCHRP Project 9-25: Requirements for Voids in
Mineral Aggregate for Superpave Mixtures, Interim Report to the
National Cooperative Highway Research Program, Advanced Asphalt
Technologies, LLC, September 2001.

Christensen, D. W., and D. A. Anderson, Interpretation of Dynamic
Mechanical Test Data for Paving Grade Asphalt, Journal of the
Association of Asphalt Paving Technologists, Vol. 61, 1992, pp. 67-98

Hirsch, T. J., Proceedings of the American Concrete Institute, Vol. 59,
1962, p. 427.

Lytton, R. L., J. Uzan, E. G. Fernando, R. Roque, D. Hiltunen, and S. M.
Stoffels, Development and Validation of Performance Prediction Models
and Specifications for Asphalt Binders and Paving Mixes, Report SHRP-
A-357, Washington, D.C.: Strategic Highway Research Program, 1993.

Nichols, R., Composite Construction Materials Handbook, Englewood
Cliffs, N. J.: Prentice-Hall, Inc., 1976, pp. 259-262.

Pellinen, T. K., Investigation of the Use of Dynamic Modulus as an
Indicator of Hot-Mix Asphalt Performance, A Dissertation Presented in
Partial Fulfillment of the Requirements for the Degree Doctor of
Philosophy, Arizona State University, May 2001, 803 pp.

Shashidhar, N., S. P. Needham, B. H. Chollar, and P. Romero,
Prediction of the Performance of Mineral Fillers in SMA, Journal of
the Association of Asphalt Paving Technologists, Vol. 68, 1999, p. 222.

Tayebali, A. A., J. A. Deacon, and C. L. Monismith, Development and
Evaluation of Surrogate Fatigue Models for SHRP A-003A Abridged
Mix Design Procedure, Journal of the Association of Asphalt Paving
Technologists, Vol. 64, 1995, pp. 340-364.

Witczak, M. W., M. Bari, and M. M. Quayum, Sensitivity of Simple
Performance Test Dynamic Modulus |E*|, NCHRP 9-19 Subtask C4b
Report, Tempe, AZ: Arizona State University, Department of Civil and
Environmental Engineering, December 2001.

Potrebbero piacerti anche