Geo-China 2016 GSP 266 59

Predicting the Dynamic Modulus of Asphalt Concrete by Binder DSR Testing

Downloaded from ascelibrary.org by Tufts University on 10/19/16. Copyright ASCE. For personal use only; all rights reserved.

A. S. M. Asifur Rahman1; Umme A. Mannan2; and Rafiqul A. Tarefder3

1
Ph.D. Student, Dept. of Civil Engineering, Univ. of New Mexico, MSC01 1070, 1 University of New
Mexico, Albuquerque, NM 87131. E-mail: arahman@unm.edu
2
Ph.D. Student, Dept. of Civil Engineering, Univ. of New Mexico, MSC01 1070, 1 University of New
Mexico, Albuquerque, NM 87131. E-mail: uam@unm.edu
3
Professor, Dept. of Civil Engineering, Univ. of New Mexico, MSC01 1070, 1 University of New
Mexico, Albuquerque, NM 87131. E-mail: tarefder@unm.edu

Abstract: This study proposes a completely new regression-based predictive model to

estimate dynamic modulus (|E*|) of Asphalt Concrete (AC) from the dynamic shear
modulus (|Gb*|) and phase angle (δb) of the asphalt binder used in the AC mix. Other
parameters related to the aggregate gradation and volumetric are also incorporated in
the model. In this study, a total of 10 AC mixes with 5 binders having different
Performance Grades (PG) and sources were collected from the manufacturing plants.
The AC mixes were compacted and cored to cylindrical specimens. After that, the
samples were tested in the laboratory for |E*| at different temperatures and loading
frequencies. The collected binders were tested for |Gb*| and δb using Dynamic Shear
Rheometer (DSR). The statistical comparison shows that a fairly accurate estimation
of |E*| can be found by using this new |E*| predictive model.

INTRODUCTION
The complex modulus (E*) is a complex number that relates applied stress to the
recoverable strain for a linear viscoelastic material (e.g., asphalt concrete) subjected to
sinusoidal loading. The two viscoelastic material properties that can be determined
from a complex modulus (E*) test are the dynamic modulus (|E*|) and the phase angle
(φ). The absolute value of complex modulus is referred to as |E*|. The phase angle, φ
is the phase difference between applied stress and measured strain response (Meyers
and Chawla 1999). In case of Asphalt Concrete (AC), laboratory determination of |E*|
function involves application of cyclic load as a function of frequency and
temperature. The resulting strain is then measured. In viscoelastic material like
asphalt, the behavior is somewhere in between that of purely elastic and purely
viscous materials, exhibiting some phase lag less than that for purely viscous materials
(i.e., 90°) and greater than purely elastic materials (i.e., 0°).
The |E*| of AC depends on many factors such as aggregate gradation, binder
property, mix volumetric, etc. Several empirical models are available in the literature
addressing these factors to determine |E*| of AC; of which the most commonly used
are the viscosity (η) based Witczak model, the binder shear modulus (|Gb*|) based

© ASCE

Geo-China 2016
 Geo-China 2016 GSP 266 60

Witczak model, and the Hirsch model (Weldegiorgis 2014). The η-based Witczak
model is the primary |E*| predictive model in the recently developed pavement design
software AASHTOWare-ME. This model uses the η of binder as the main input
parameter to capture the effect of binder, the aggregate gradation, the air void content,
and the asphalt content on |E*| of AC mix (NCHRP 2004).
The performance of the η-based Witczak model for predicting |E*| of AC was
evaluated by several researchers. Clyne et al. (2003), Christensen et al. (2003), Tran
and Hall (2005), and Mohammad et al. (2005) reported that the η-based Witczak
model produces slightly less value of |E*| compared to the tested data. Schwartz
Downloaded from ascelibrary.org by Tufts University on 10/19/16. Copyright ASCE. For personal use only; all rights reserved.

(2005) concluded that the η-based Witczak model under-predicts |E*| of AC. On the
other hand, Birgisson et al. (2005) found an over prediction of |E*| value by the η-
based Witczak model. Kim et al. (2005) reported that the η-based Witczak equation
predicts better at low temperature.
The |Gb*|-based Witczak model (Bari and Witczak 2006) was developed using the
Bari (2005) database of 7400 measured |E*| values obtained from 346 different Hot-
Mix Asphalt (HMA) mixes. The data used to develop the earlier version of the model
are included in this expanded database. However, to convert conventional viscosity
temperature susceptibility parameters A and VTS to |Gb*| and δb, empirical models are
also provided, which is because at that time, Superpave binder characterization were
unavailable (Ceylan et al. 2009). Some issues have been raised regarding the use of
inconsistent treatment of loading frequency in case of AC and binder (Christiansen
2006). Singh et al. (2011) reported that the accuracy of the |Gb*|-based Witczak model
to be poor when compared to other available |E*| models in the literature. El-Badawy
et al. (2012) concluded that the |Gb*|-based |E*| predictive model produced less
accurate and relatively higher biased estimates of |E*| than the η-based Witczak
model.
Christensen et al. (2003) investigated several versions of the Hirsch model and
found the most effective one to be the simplest one in which |E*| is directly estimated
from binder |Gb*|, voids in mineral aggregate (VMA), and voids filled with asphalt
binder (VFA). Singh et al. (2011) also investigated Christensen’s proposed Hirsch
model for predicting |E*| of AC and found that the predicted |E*| values are dispersed
around the Line of Equality (LOE) while compared to the measured |E*|, indicating
that the model exhibits significant error. Bari and Witczak (2006), Obulareddy (2006),
King et al. (2005), and Ceylan et al. (2008) reported that the Hirsch model under-
predicts |E*| compared to the tested data. However, the fundamental weakness of the
model includes strong dependence on volumetric parameters, particularly under low
air void and VFA conditions, and questions regarding the ability of the |Gb*|
parameters to account for the possible beneficial effects of modifiers (Al-Khateeb et
al. 2006, Soleymani et al. 2004).

OBJECTIVES
The objective of this current study is to develop an alternative, more accurate,
regression-based |E*| predictive model for the AC mixtures and binders typically used
by the state of New Mexico. The model should be capable of estimating changes in the
|E*| of AC mixtures as a function of aggregate gradation parameters, mixture
volumetric, binder properties, temperature and loading frequency. Other objectives

© ASCE

Geo-China 2016
 Geo-China 2016 GSP 266 61

include: collection of plant produced asphalt mixtures and binders, laboratory testing
for |E*| at different temperatures and loading frequencies, laboratory testing for |Gb*|
and δb at different temperatures and frequencies of loading for the binders used in the
AC mixtures. A non-linear regression optimization is used for the model development.
Also, for the incorporation of the aggregate gradation in the model, two universally
known gradation parameters, called the aggregate uniformity coefficient (Cu) and the
fineness modulus (Fm) are introduced instead of using directly the percentage of
material retained or passing on a particular sieve, which is pretty common in most of
the |E*| models currently in practice.
Downloaded from ascelibrary.org by Tufts University on 10/19/16. Copyright ASCE. For personal use only; all rights reserved.

MATERIALS
A total of ten loose AC mixtures (referred to as Mix-1 through Mix-10) with four
asphalt binders of different PG grades were collected from the production plants and
paving sites for laboratory testing. A summary of the collected AC mixtures and
binders are presented in Table 1. Liquid asphalt binders used in the AC mixtures were
collected in cans from the asphalt mixing plant.

Table 1. Summary of Collected Asphalt Concrete Mixtures and Asphalt Binders

Mix ID Superpave Nominal Maximum Asphalt Binder Type of Aggregate
Gradation Aggregate Size Performance Grade Material
(NMAS), mm (PG)
Mix 1 SP IV 12.5 PG 64-22 Alluvial Limestone
Mix 2 SP III 19.0 PG 64-28 Sand & Gravel
Mix 3 SP III 19.0 PG 76-28 Dacite
Mix 4 SP III 19.0 PG 76-28 Basalt
Mix 5 SP III 19.0 PG 76-22 Sand & Gravel
Mix 6 SP III 19.0 PG 64-28 Basalt
Mix 7 SP III 19.0 PG 64-22 Limestone
Mix 8 SP III 19.0 PG 64-22 Limestone
Mix 9 SP III 19.0 PG 64-22 River Deposits
Mix 10 SP III 19.0 PG 64-28 River Deposits

PREPARATION OF TEST SPECIMENS

A Superpave® gyratory compactor was used to compact loose AC mixtures.
Cylindrical AC cores of 150 mm in diameter and 170 mm in height were compacted.
The target air void was set at 5.5±0.5% for the finished specimens to avoid possible
deviation in test results due to large air void variation. Several trial mixes were
compacted at the beginning with different weights of loose AC material. The required
number of gyrations to reach 170 mm overall height was then noted for each trial AC
cores. The compacted samples were then core-drilled and sawed to finished specimens
of diameter 100 mm and of height 150 mm. The theoretical maximum specific gravity
(Gmm) was determined by AASHTO T 209 (2012). The bulk specific gravity (Gmb) was
determined according to the AASHTO T 166 (2012) protocol. The loose mix weights
corresponding to the air void content of 5.5±0.5% were then used to compact further
AC cores. For each AC mixture, 3 cylindrical specimens were prepared for |E*|
testing. For conducting DSR testing on binder samples, AASHTO T 240 (2009)
standard was used to conduct RTFO aging.

© ASCE

Geo-China 2016
 Geo-China 2016 GSP 266 62

LABORATORY DYNAMIC MODULUS TEST

The |E*| testing was conducted according to AASHTO T 342 (2011) protocol. All
the AC specimens were tested for |E*| at five test temperatures, 14, 40, 70, 100, and
130 °F (-10, 4.4, 21.1, 37.8, and 54.4 °C) and six loading frequencies of 25, 10, 5, 1,
0.5, and 0.1 Hz. The average of |E*| of the three specimens was then determined.

FREQUENCY SWEEP COMPLEX SHEAR MODULUS TEST

The AASHTO T 315 (2012) test standard was employed as a guideline for
conducting dynamic shear modulus, |Gb*| and δb testing on the binder samples. |Gb*|
Downloaded from ascelibrary.org by Tufts University on 10/19/16. Copyright ASCE. For personal use only; all rights reserved.

with δb tests were conducted at several test temperatures. These are 130 °F, 115 °F,
100 °F, 85 °F, 70 °F, 55 °F, and 40 °F. Instead of testing only at one frequency of 10
rad/sec as required in AASHTO T 315, |Gb*| tests were conducted at 11 frequencies
ranging from 1.0 to 100 rad/sec for each temperature. Two different sample sizes were
used for conducting |Gb*| tests. Samples with 25 mm in diameter and 1 mm in height
were used for 130 °F and 115 °F temperatures. For other temperatures, samples with 8
mm in diameter and 2 mm in height were used. Different sizes of samples were used
because torsion force required to maintain a measurable strain level of a 25 mm
sample at lower temperatures exceeds the equipment capacity. The |Gb*| tests were
conducted in a strain controlled mechanism. The applied strain level was 1.0%.

THE DYNAMIC MODULUS PREDICTIVE MODEL

A nonlinear regression approach was used to develop the new |E*| predictive model.
At first, the time-temperature superposition principal (TTSP) was applied to develop
the average |E*| mastercurves for the AC mixtures and the |Gb*| mastercurves for the
associated binders at 70 °F reference temperature. After application of TTSP, both of
the functions follow sigmoidal shape; the following sigmoid expression was used to fit
these functions.
α
log | F (ω ) | = δ MC + β +γ log (ωr )
(1)
1+ e
In Eq. (1), F(ω) is the material functions, i.e. |E*(ω)| for AC mixtures, and |Gb*(ω)|
for the binder. Note that, both of these functions are kept in angular frequency space,
designated by ω. The relationship between the angular frequency in rad/sec and
ordinary frequency (f) in Hz is given in Eq. (2).
ω = 2π f (2)
In current practice, |E*| is determined in ordinary frequency space and |Gb*| is
determined in angular frequency space. Therefore, to study the effect of different
variables in the proposed model, it was necessary to convert the ordinary frequency
space to angular frequency space in case of |E*| functions of the mixtures. The other
parameters in Eq. (1), such as, ωr is the reduced angular frequency of loading, and α,
β, γ, δMC are the fitting parameters.
The variables affecting | E*(ω)| of AC mix considered in this study are: the fineness
modulus (Fm), and uniformity coefficient (Cu) for the aggregate used; effective percent
volume of asphalt binder (Vbeff), percent air void (Va), the dynamic shear modulus
(|Gb*|) and the associated phase angle (δb) of the binder at a given loading rate. The
definitions of Fm, and Cu can be given as:

© ASCE

Geo-China 2016
 Geo-China 2016 GSP 266 63

 CPR i
Fm = i =1
(3)
100
D
Cu = 60 (4)
D10
where, CPRi is the cumulative percentage of aggregate retained at ith sieve, D60 is the
sieve size corresponding to 60% material passing, D10 is the sieve size corresponding
to 10% material passing and n is the number of sieves used in the analysis.
Downloaded from ascelibrary.org by Tufts University on 10/19/16. Copyright ASCE. For personal use only; all rights reserved.

A two-step modeling approach was adopted while developing the |E*| predictive
model. In the first step, the first seven AC mixtures and associated binders were used
to develop the model, and the modulus data of remaining AC mixtures and binders
were tested. In the second step, all 10-AC mixtures were used to obtain the final form
of the predictive model. The expression given in Eq. (5) is the final form of the model.
Note that the |E*| is kept at angular frequency space (|E*(ω)|) so that the angular
loading frequency associated with a certain pair of |Gb*| and δb can be readily used in
the equation to find out the |E*(ω)| of the AC mix for that particular angular frequency
of loading. In Eq. (5), both the |E*(ω)| and |Gb*| are in pound per square inch (psi)
unit.
−1.775
 Vbeff 
log E * (ω ) = −1.548 + 0.0155 (Fm ) + (Cu ) +  
0.35
V +V 
 beff a 

 Vbeff  (5)
0.75 (Fm ) − 0.495 (Cu ) + 3.875  
0.6 0.35
V +V 
 beff a 
+ − 0.69 − 0.47 log Gb * + 0.59 log δ b
1+ e

MODEL EVALUATION AND DISCUSSION

The development of a regression-based model greatly depends on the statistical
analysis and optimization process used. The fundamental aim of this kind of model
development process is to reduce the error from the prediction by comparing the
predicted data with the observed data for the identical input variables. Optimization
process involves the determination of regression coefficients in such a way that the
developed equation provides minimum error when the predicted and observed data are
compared. To determine the level of accuracy of the model, a statistical evaluation
called “goodness of fit” is used. To determine “goodness of fit”, the estimated values
by the model are compared with the tested or observed values at the same input
conditions. Generally, two statistical parameters are required to be evaluated to
determine the “goodness of fit” of the model. The first one is the coefficient of
determination (R2), and the second one is the ratio of the standard error (Se) to the
standard deviation (Sy). The mathematical form of R2, Se, and Sy can be presented by
the expressions given in Eq. (6), Eq. (7), and Eq. (8).
2
R = 1−
(n − p − 1)S e
2

(6)
(n − 1)S y 2

© ASCE

Geo-China 2016
 Geo-China 2016 GSP 266 64

n
1
Se = 
(n − p − 1) 1
(xˆi − xi )2 (7)

n
1
Se = 
(n − 1) 1
(xi − x )2 (8)

In Eqs. (6), (7), and (8), x̂i are the predicted data, xi are the observed data, x is the
average of the observed data, n is the number of data points used in the model, and p is
the number of fitting parameters used in the model. A relatively good predictive model
Downloaded from ascelibrary.org by Tufts University on 10/19/16. Copyright ASCE. For personal use only; all rights reserved.

would have a higher R2, close to 1 and a smaller Se/Sy.

The laboratory tested |E*(ω)| at different angular frequencies at a given temperature,
which were fitted by the power law of the following form:
| E * (ω ) |Ti = p1 (ω ) p2 (9)
where, |E*(ω)|Ti are the |E*(ω)| at a given temperature, p1 and p2 are the fitting
parameters. The fitted equations were than used to find the |E*(ω)| corresponding to
the angular frequencies associated with the tested |Gb*| and δb of the binder. This is
because of that the |E*(ω)| tests were conducted at different angular frequencies than
that of in the |Gb*| tests. Therefore, it was necessary to find the |E*(ω)| at the angular
frequencies same as in |Gb*| tests. As the tested |E*(ω)| data was fitted by the power
law, the calculated |E*(ω)| from this fitted curve is considered actual tested data.
The goodness of fit of the proposed |E*(ω)| model in Eq. (5) was evaluated in two
ways, in normal (arithmetic) scale and in logarithmic scale. The developed |E*(ω)|
model has a fairly good coefficient of determination (R2 = 0.8687) and a very small
Se/Sy (Se/Sy = 0.3648) in arithmetic scale. Again, in logarithmic scale these are: R2 =
0.8990, and Se/Sy = 0.3199, which are fairly good for this type of models where
numerous complexities are involved. Figures 1 and 2 show the observed |E*(ω)| data
versus the |E*(ω)| data predicted from the model in arithmetic and logarithmic scale
respectively. Both of the plots show that all the |E*(ω)| data points are around the line
of equality (LOE) without any significant bias. Therefore, it can be said that the
proposed |E*| model based on |Gb*| gives a fairly good prediction of |E*(ω)| of the AC
mixtures under concern.
8E+06
Line of
R2 = 0.8687 Equality
Se = 566 ksi (LOE)
Predicted |E*|, psi

6E+06 Sy = 1551 ksi

Se/Sy = 0.3648
4E+06

2E+06

0E+00
0E+00 2E+06 4E+06 6E+06 8E+06
Laboratory Tested |E*|, psi
(Actual test data fitted by power law)
Fig. 1. Laboratory tested |E*| versus predicted |E*| plot in arithmetic scale.

© ASCE

Geo-China 2016
 Geo-China 2016 GSP 266 65

8.00
R2 = 0.8990

Predicted log|E*|, psi

Se = 0.1719 Line of
7.00
Sy = 0.5374 Equality
Se/Sy = 0.3199 (LOE)
6.00

5.00
Downloaded from ascelibrary.org by Tufts University on 10/19/16. Copyright ASCE. For personal use only; all rights reserved.

4.00
4.00 5.00 6.00 7.00 8.00
Laboratory Tested log|E*|, psi
(Actual test data fitted by power law)
Fig. 2. Laboratory tested |E*| versus predicted |E*| plot in logarithmic scale.
CONCLUSIONS
The following conclusions can be drawn from this study:
1. A new dynamic modulus (|E*(ω)|) predictive model is developed in this study
based on observed data from 10 asphalt concrete mixtures typically found in New
Mexico.
2. The developed model uses two fundamental aggregate gradation parameters: the
fineness modulus and the uniformity coefficient, mix volumetric parameters (air void
content and effective binder volume), and binder rheological parameters (shear
modulus and phase angle) as direct input.
3. The developed |E*(ω)| model possesses fairly good statistics, considering
goodness of fit of the model.

REFERENCES
Al-Khateeb, G., Shenoy, A., Gibson, N., and Harman, T. (2006). “A new simplistic
model for dynamic modulus predictions of asphalt paving mixtures.” J. of the
Assoc. of Asphalt Paving Technologists, 75.
Bari, J. (2005). Development of a new revised version of the Witczak E* predictive
models for hot mix asphalt mixtures. Ph.D. Dissertation, Arizona State University,
Tempe, AZ.
Bari, J., and Witczak, M.W. (2006). “Development of a new revised version of the
Witczak E* predictive model for hot mix asphalt mixtures (with discussion).” J. of
the Assoc. of Asphalt Paving Technologists, 75: 381–423.
Birgisson, B., Sholar, G., and Roque, R. (2005). “Evaluation of a predicted dynamic
modulus for Florida mixtures.” Transp. Research Rec.: J. of the Transp. Res.
Board, 1929: 200–207.
Christiansen, D. (2006). “Published discussion to ‘Bari and Witczak’.” J. of the Assoc.
of Asphalt Paving Technologists, 75: 422–423.
Christensen, Jr., D.W., Pellinen, T., and Bonaquist, R.F. (2003). “Hirsch model for
estimating the modulus of asphalt concrete.” J. of the Assoc. of Asphalt Paving
Technologists, 72: 97–121.

© ASCE

Geo-China 2016
 Geo-China 2016 GSP 266 66

Ceylan, H., Gopalakrishnan, K., and Kim, S. (2008). “Advanced approaches to hot-
mix asphalt dynamic modulus prediction.” Canadian J. Civil Engin., 35(7): 699–
707.
Ceylan, H., Schwartz, C. W., Kim, S., & Gopalakrishnan, K. (2009). “Accuracy of
predictive models for dynamic modulus of hot-mix asphalt.” Journal of Materials
in Civil Engineering, 21(6), 286–293.
Clyne, T.R., Li, X., Marasteanu, M.O., and Skok, E.L. (2003). Dynamic and resilient
modulus of Mn/DOT asphalt mixtures (No. MN/RC-2003-09).
El-Badawy, S., Bayomy, F., and Awed, A. (2012). “Performance of MEPDG dynamic
Downloaded from ascelibrary.org by Tufts University on 10/19/16. Copyright ASCE. For personal use only; all rights reserved.

modulus predictive models for asphalt concrete mixtures: local calibration for
Idaho.” J. of Mater. Civil Engin., 24(11): 1412–1421.
Kim, Y.R., King, M., and Momen, M. (2005). “Typical dynamic moduli values of Hot
Mix Asphalt in North Carolina and their prediction.” 84th Annual Meeting of the
TRB, Paper No. 05-2568, Washington D.C.
King, M., Momen, M., and Kim, Y.R. (2005). “Typical dynamic modulus values of
hot-mix asphalt in North Carolina and their prediction.” Presented at 84th Annual
Meeting of the Transportation Research Board, Washington, D.C.
Meyers, M.A., and Chawla, K.K. (1999). Mechanical Behavior of Materials, Prentice
Hall, pp. 98–103.
MEPDG (2008). Mechanistic Empirical Pavement Design Guide. A manual of
practice. American Association of State Highway and Transportation Officials.
Mohammad, L.N., Wu, Z., Myers, L., Cooper, S., and Abadie, C. (2005). “A practical
look at simple performance tests: Louisiana’s experience.” J. of the Assoc. of
Asphalt Paving Tech., Vol. 74: 557–600.
NCHRP (2004). Guide for Mechanistic-Empirical Design of new-rehabilitated
pavement structures. Report 1-37A, National Cooperative Highway Research
Program (NCHRP), Washington, D.C.
Obulareddy, S. (2006). Fundamental characterization of Louisiana HMA mixtures for
the 2002 mechanistic-empirical design guide. MS Thesis, Louisiana State
University, Baton Rouge.
Schwartz, C.W. (2005). “Evaluation of the Witczak dynamic modulus prediction
model.” 84th Annual Meeting of the Transportation Research Board, Paper No.
05-2112, Washington D.C.
Singh, D., Zaman, M., and Commuri, S. (2011). “Evaluation of predictive models for
estimating dynamic modulus of hot-mix asphalt in Oklahoma.” Trans. Re. Rec.: J.
of the TRB, Vol. 2210: 57–72.
Soleymani, H., Zhai, H., & Bahia, H. (2004). “Role of modified binders in rheology
and damage resistance behavior of asphalt mixtures.” Transp. Res. Rec.: J. of the
TRB, Vol. 1875: 70–79.
Tran, N.H., and Hall, K.D. (2005). “Evaluating the Predictive Equation in
Determining Dynamic Moduli of Typical Asphalt Mixtures Used in Arkansas.” J.
of the Assoc. of Asphalt Paving Technologists, Vol. 74: 1–17.
Weldegiorgis, M. (2014). On dynamic modulus of asphalt concrete for moisture
damage. Doctoral Dissertation, Department of Civil Engineering, University of
New Mexico, Albuquerque, NM, USA.

© ASCE

Geo-China 2016