Sei sulla pagina 1di 20

Micromechanical Modeling of Asphalt

Concrete Uniaxial Creep Using


the Discrete Element Method

Habtamu Melese Zelelew* — Athanassios Tom Papagiannakis**


Downloaded by [Federal Highway Admin Research] at 07:39 01 August 2012

* University of Minnesota Duluth


Duluth, MN 55812
(218) 726-8427
hzelelew@d.umn.edu
** Civil and Environmental Engineering
University of Texas-San Antonio
San Antonio TX 78249-0668
(210) 458-7071
at.papagiannakis@utsa.edu

ABSTRACT. This paper describes a methodology for modeling the creep behavior of Asphalt
Concretes (ACs) under uniaxial loading. The approach involves capturing the microstructure
of asphalt concretes using X-Ray Computed Tomography (CT) and using the Discrete
Element Method (DEM) to describe the microstructure. The viscoelastic properties of mastics
were characterized by fitting the Burger model to Dynamic Shear Rheometer (DSR) data. An
automated digital image processing technique called Volumetric-based Global Minima
(VGM) thresholding algorithm was utilized to process the X-Ray CT images of AC cores. The
DEM simulations results were compared favorably to experimental uniaxial creep data. It
was found that the models underestimated the creep compliance of the mixtures in the
primary creep stage and predicted quite accurately in the secondary creep stage. The
maximum absolute error observed for the slope and intercept of this region was not higher
than 2.6%. Although the models predicted flow time, no meaningful comparison could be
made with experimental flow times.
KEYWORDS: Asphalt Concrete, Uniaxial Creep, Discrete Element.

DOI:10.3166/RMPD.11.613-632 © 2010 Lavoisier, Paris

Road Materials and Pavement Design. Volume 11 – No. 3/2010, pages 613 to 632
614 Road Materials and Pavement Design. Volume 11 – No. 3/2010

1. Introduction

Asphalt Concrete (AC) mixtures are uniquely complex heterogeneous materials


composed of air voids, mastic and aggregates. Mastics are blends of asphalt binder
and fines, typically considered as particles passing sieve No. 200 (i.e., sizes finer
than 75 microns). The distribution of these three phases and their interaction defines
the mechanical properties of ACs and contributes significantly to their rutting
resistance potential, load carrying capacity and durability.
Pavement rutting consists of surface depressions along the wheel paths caused by
Downloaded by [Federal Highway Admin Research] at 07:39 01 August 2012

the plastic deformation of the AC and the underlying granular layers under traffic
loads. The plastic deformation in the asphalt concrete layers is a function of the
properties of the mastic and the physical properties of aggregate (e.g., shape (or
form), angularity (or roundness), and surface texture) and governs the performance
of AC pavements. The properties of the mastic depend on the properties of the
asphalt binder, which depend on temperature and rate of loading. Recently, Zelelew
and Papagiannakis (2010) investigated the effects of rheological properties of
mastics on creep performance of ACs. The Bingham rheological model was used to
estimate the yield stress of asphalt mastic samples. The study showed that mastic
samples with higher yield stress values provided better deformation resistance AC
mixtures. Several studies have also shown that the permanent deformation behavior
of AC mixtures is highly dependent on aggregates characteristics (Sousa et al.,
1991; Masad et al., 2002; Tashman et al., 2002). Therefore, proper characterization
of the resistance of ACs to permanent deformation needs to account for both the
binder properties and the aggregate microstructure.
Continuum-based Finite Element Method (FEM) and micromechanical-based
Discrete Element Method (DEM) modeling approaches are the two commonly used
numerical simulation techniques to model ACs. Moreover, there have been efforts to
model the behavior of AC mixtures using hybrid modeling schemes, combining
these two modeling approaches. Recently, the FEM-DEM modeling and simulation
techniques have been implemented to characterize the AC mixture properties under
a range of temperature and laboratory testing condition (Dai and You, 2007).
The AC behavior is dominated by the interaction between mastics and distinct
aggregate particles. The role of aggregate structures in ACs was investigated and
demonstrated that AC mixtures behave as granular materials (Shashidhar, 2000).
DEM is suited to describe particle contact interaction (Cundall and Strack, 1979). In
the past, the application of DEM was restricted by the lack of sufficient memory and
computational power. This is no longer a problem however, which makes DEM an
attractive tool for simulating the AC microstructure under large deformations.
Previous studies on modeling the AC mixture behavior using the micromechanical
modeling approach include Rothenburg et al. (1992), Chang and Meegoda (1997),
Buttlar and You (2001), You and Buttlar (2002), You and Buttlar (2004), Abbas et
al. (2005), and Collop et al. (2006).
Micromechanical Modeling of Asphalt Concrete Uniaxial Creep 615

Rothenburg et al. (1992) proposed a micromechanical discrete model for AC


pavement rutting. In this model, the AC was represented by a set of discrete elastic
elements bound by a linear viscoelastic binder. It was demonstrated that the
mechanical response of ACs is highly dependent on the characteristics of the
aggregate phase. Simulation of creep tests showed that the steady-state creep
properties are largely controlled by frictional inter-granular contact, rather than
cohesive inter-granular contact.
Another innovative DEM model was proposed, by modifying the TRUBAL
program (Cundall, 1971), to describe the interaction of aggregate-to-aggregate and
Downloaded by [Federal Highway Admin Research] at 07:39 01 August 2012

asphalt-to-aggregate contacts (Chang and Meegoda, 1997). A mechanistic model


was used to simulate the viscoelastic behavior of asphalt binder using linear
viscoelastic elements. Based on the mechanical responses and comparisons with
experimental results, a Burger model was found to be the most promising asphalt
binder behavior model. Recently, researchers developed simulation techniques by
extending the conventional DEM analysis to account for the mastic and aggregate
phases and model the AC microstructure using the Microfabric Distinct Element
Method (MDEM) (Buttlar and You, 2001). This study described the AC
microstructure using clusters of circular particles to simulate the behavior of AC
mixtures under Indirect Tension Test (IDT) conditions. This technique was also used
to examine the importance of aggregate-to-aggregate contact on the uniaxial
compression of mastics and coarse grained mixtures (You and Buttlar, 2002).
The literature to date provides valuable insight into the modeling of the
microstructure of asphalt concretes. A number of challenges remain, namely the
characterization and modeling of the contact laws between mastic and aggregate
particles, particularly as the limits of linear viscoelasticity are reached and beyond at
the point were the contacts are broken. The paper at hand addresses some of these
modeling challenges.
Rothenburg et al. (1992) proposed a micromechanical discrete model for AC
pavement rutting. In this model, the AC was represented by a set of discrete elastic
elements bound by a linear viscoelastic binder. It was demonstrated that the
mechanical response of ACs is highly dependent on the characteristics of the
aggregate phase. Simulation of creep tests showed that the steady-state creep
properties are largely controlled by frictional inter-granular contact, rather than
cohesive inter-granular contact.
Another innovative DEM model was proposed, by modifying the TRUBAL
program (Cundall, 1971), to describe the interaction of aggregate-to-aggregate and
asphalt-to-aggregate contacts (Chang and Meegoda, 1997). A mechanistic model
was used to simulate the viscoelastic behavior of asphalt binder using linear
viscoelastic elements. Based on the mechanical responses and comparisons with
experimental results, a Burger model was found to be the most promising asphalt
binder behavior model. Recently, researchers developed simulation techniques by
extending the conventional DEM analysis to account for the mastic and aggregate
616 Road Materials and Pavement Design. Volume 11 – No. 3/2010

phases and model the AC microstructure using the Microfabric Distinct Element
Method (MDEM) (Buttlar and You, 2001). This study described the AC
microstructure using clusters of circular particles to simulate the behavior of AC
mixtures under Indirect Tension Test (IDT) conditions. This technique was also used
to examine the importance of aggregate-to-aggregate contact on the uniaxial
compression of mastics and coarse grained mixtures (You and Buttlar, 2002).
The literature to date provides valuable insight into the modeling of the
microstructure of asphalt concretes. A number of challenges remain, namely the
characterization and modeling of the contact laws between mastic and aggregate
Downloaded by [Federal Highway Admin Research] at 07:39 01 August 2012

particles, particularly as the limits of linear viscoelasticity are reached and beyond at
the point were the contacts are broken. The paper at hand addresses some of these
modeling challenges.

2. Objective

The objective of this paper is to present a DEM-based approach for simulating


the creep behavior of AC mixtures and compare these predictions to experimental
creep data obtained under uniaxial loading. The AC microstructure of the test cores
is captured using X-Ray Computed Tomography (CT) imaging techniques. The
mastic is modeled to the limits of its linear viscoelastic behavior using Burger
models fitted to Dynamic Shear Rheometer (DSR) data.

3. Experimental data

3.1. Creep tests on asphalt concrete mixtures

The laboratory creep data analyzed was generated by a Texas DOT study
conducted by the University of Texas-El Paso (Alvarado et al., 2007). One of the
mix design in this report is analyzed here, namely a Coarse Matrix High Binder type
C (CMHB-C). Three types of mixtures were prepared involving aggregates from
three different sources, namely a Hard Limestone (HL), a Soft Limestone (SL), and
a Granite (G). A PG 76-22 binder was used for all mixtures. Table 1 summarizes the
volumetric data for the cores produced. The Superpave gyratory compacted
specimens, which measured 150 mm in diameter by 165 mm in height were drilled
to 100 mm diameter and trimmed to a height of 150 mm. After trimming, the cores
were scanned using the X-Ray CT facility at Texas A&M University. Subsequently,
the cores were subjected to uniaxial creep testing under a constant vertical stress of
207 kPa at a chamber temperature of 60°C.
Micromechanical Modeling of Asphalt Concrete Uniaxial Creep 617

Table 1. AC mixture design parameters; Coarse Matrix High Binder type C


(CMHB-C) Mixture (Alvarado et al., 2007)

Hard
Soft Limestone
Parameter Limestone Granite (G)
(HL)
(HL)
Binder Grade PG 76-22
Binder Content, % 4.2 5.3 5.8
Downloaded by [Federal Highway Admin Research] at 07:39 01 August 2012

Sieve Size/Sieve No. (mm) Percent Passing (%)

1 (25.0) 100 100 100

3/4 (18.75) 99 99 99

1/2 (12.5) 78.5 78.5 78.5

3/8 (9.38) 60 60 60

No. 4 (4.75) 37.5 37.5 37.5

No. 8 (2.36) 22 22 22

No. 16 (1.18) 16 16 16

No. 30 (0.600) - - -

No. 50 (0.300) - - -

No. 200 (0.075) 7 7 7

Maximum Specific Gravity 2.554 2.471 2.450

Aggregate Bulk Specific Gravity 2.696 2.601 2.587

Binder Specific Gravity 1.02

Air Voids at Ndesign = 100 (%) 7.3 6.9 7.0

VMA at Ndesign = 100 (%) 12.7 13.7 14.3

VFA at Ndesign = 100 (%) 70.2 69.7 72.5

Effective Binder Content (%) 3.7 4.1 4.5

Dust Proportion (%) 1.7 1.3 1.2


618 Road Materials and Pavement Design. Volume 11 – No. 3/2010

3.2. AC microstructure characterization

Recently, advances imaging techniques have gained popularity in characterizing


the microstructure of ACs. Several researchers implemented Digital Image
Processing (DIP) techniques to study the microstructure of AC mixtures. These
techniques have been used to obtain quantitative information about the air void,
mastic and aggregate distribution, shape of the aggregate particles, aggregate
orientation, aggregate gradation, aggregates contacts, aggregate segregation and so
on. Despite these developments, identifying the three phases in AC X-Ray CT
images has been done to a large extent subjectively. This involves establishing the
Downloaded by [Federal Highway Admin Research] at 07:39 01 August 2012

grey level thresholds that separate aggregate from mastic and mastic from air.
X-Ray CT is a non-destructive advanced imaging technique that generates two-
and three-dimensional high resolution images with the capability of capturing the
details of the microstructure. This technique has been used by several researchers to
characterize different properties of AC materials. Recently, it is used to effectively
quantify air void distribution, aggregate orientation, and segregation (Masad et al.,
1999; Shashidhar, 1999; Masad et al., 2002; Tashman et al., 2002; and Zelelew and
Papagiannakis, 2008b).
The AC cores were scanned perpendicular to their vertical axis at 1 mm distance
intervals to yield 148 slices per core, ignoring the top and bottom slices. The
procedures used for capturing the AC images using X-Ray CT are well documented
by (Masad et al., 2002). Recently, Zelelew and Papagiannakis (2008a) developed an
automated DIP technique called Volumetric-based Global Minima (VGM)
thresholding algorithm. VGM was utilized to process the horizontally sliced X-Ray
CT images. The method is based on identifying gray level boundary thresholds
between air, mastic and aggregate phases with reference to volumetric information.
VGM has three inter-related stages. The first stage involves image preprocessing for
contrast enhancement and noise removal. The second stage is the main thresholding
routine accepting as input the enhanced images of the first stage and volumetric
information for the AC. It consists of two components, namely volumetrics-driven
thresholding and three-dimensional representation which yields two-dimensional
images through sectioning. The third stage further enhances particle separation
through edge detection and image segmentation techniques.
An example of the resulting two-dimensional rectangular sections produced by
combining from the circular core sections for mixture HL CMHB-C is given in
Figure 1a. The resolution of this image is 195Pm per pixel, which does not allow
detecting particles larger than roughly particles passing sieve No. 70. Figure 1b
shows the corresponding air phase. These figures reveal that there is a larger
concentration of air voids around the sample periphery than in the middle, despite
the fact that samples were drilled. This agrees with observations from literature
(Masad et al., 1999; Masad et al., 2002). Figure 1c and 1d show the mastic and
aggregate phases, respectively. The distribution of the mastic and aggregate phases
in these figures appears to be random, which is consistent with earlier findings
Micromechanical Modeling of Asphalt Concrete Uniaxial Creep 619

(Zelelew and Papagiannakis, 2008b). Overall, the VGM technique was shown to
significantly improve the quality of the raw X-Ray CT images and render them
ready for direct input to numerical modeling and simulation.
Downloaded by [Federal Highway Admin Research] at 07:39 01 August 2012

(a) (b)

(c) (d)
Figure 1. Representation of AC rectangular sections (a) contrast enhanced and de-
noised image (b) air void phase in white, (c) mastic phase in white, and (d)
aggregate phase in white; mixture HL CMHB-C
620 Road Materials and Pavement Design. Volume 11 – No. 3/2010

3.3. Mastic viscoelastic models

The viscoelastic rheological models can accurately describe the behavior of


asphalt binders and mastics across a wide range of shear strain rates, shear stress
levels and temperatures. The phenomenological viscoelastic behavior of mastics can
be characterized by combining Hooke’s spring and Newton’s dashpot elements in
series or in parallel. The most commonly used viscoelastic rheological models for
asphalt binders and mastics is the Burger model that consists of a Maxwell and
Kelvin-Voigt element coupled in series. Several researchers have used Burger model
to describe the viscoelastic behavior of mastics (Abbas et al., 2005; Kim and Little,
Downloaded by [Federal Highway Admin Research] at 07:39 01 August 2012

2004; Yu et al., 2007).


The aggregate fines passing sieve No. 200 and binder proportions used in
preparing the mastics were obtained from the volumetric properties of mixtures
(Table 2).The PG 76-22 binder was preheated at mixing temperature of 1630C to
ensure complete fluidity and then sufficiently stirred to homogenize and remove the
air bubbles before introducing the fines. The DSR testing was conducted at testing
temperature of 600C in accordance with AASHTO T240/ASTM D2872
specifications.

Table 2. Mastic sample proportions by volume

Fine Aggregate Binder Content Mastic


Mixture ID
Content* (%) (%) (%)

HL CMHB-C 7.235 9.749 16.984

G CMHB-C 5.763 11.954 17.717

SL CMHB-C 5.374 12.956 18.330

* Fines Passing Sieve No. 200 (or Sieve Size 75 Pm); HL: Hard Limestone; G: Granite;
SL: Soft Limestone

DSR testing involved strain amplitude sweep tests for characterizing the limits of
the linear viscoelastic (LVE) range of mastics. The strain limits in this fashion were
53.5%, 42.0%, and 15.5% for the mixtures prepared with HL, G, SL aggregates,
respectively. Within these ranges, the rheological properties of mastics are
independent of the applied shear stress or strain and hence, they can be modeled
using a simple Burger model. At higher strain levels, the behavior becomes non-
Newtonian. For the purposes of this study, it was assumed that the mastic-to-mastic
bonds break in a cohesive fashion at strain levels higher than the LVE.
Additional DSR testing on mastic samples was conducted for the purpose of
fitting the Burger model. These consisted of Controlled Shear Rate (CSR) frequency
Micromechanical Modeling of Asphalt Concrete Uniaxial Creep 621

sweep tests at angular frequencies ranging from 0.01 to 100 rad/sec. These tests
were performed at strain amplitude levels equal to the LVE limits established as
described earlier. Burger models were fitted to this frequency sweep data by
minimizing the squared differences between modeled and observed storage and loss
shear moduli (Papagiannakis et al., 2002; Baumgaertel and Winter, 1989; Abbas,
2004). The constants of the Burger model obtained are shown in Table 3.

Table 3. Fitted Burger model parameters at 600C testing temperature


Downloaded by [Federal Highway Admin Research] at 07:39 01 August 2012

Burger Model Parameters

Mixture ID Maxwell Block Kelvin Block

EM (kPa) KM ( kPa.sec) EK (kPa) KK (kPa.sec)


HL CMHB-C 225.0446 45.4928 13.1130 10.5862

G CMHB-C 200.0200 30.4480 9.6397 6.3361

SL CMHB-C 149.9998 19.9957 10.9111 4.3360

HL: Hard Limestone; G: Granite; SL: Soft Limestone.

4. AC micromechanical modeling

The DEM model used in this study is the commercially available program
Particle Flow Code in Two-dimensions (PFC2D) (Itasca Consulting Group, 2003).
This two-dimensional model, treats particles as circular disks that interact through
normal and shear springs and Coulomb frictional elements. The contact forces and
displacements of a granular assembly of particles are found by tracing the
movements of the individual particles. The forces acting on any particle are
determined by its interaction with its neighboring particles. The simulation
processes in DEM involves two dependent schemes: first application of Newton’s
second law (equation of motion) to the particles and second application of a force-
displacement law at the particle contacts. The first scheme is responsible for
determining the motion of each particle arising from the contact and body forces
acting upon it. The second scheme is used to update the contact forces arising from
the relative motion at each particle contact. Furthermore, Newton’s second law is
integrated twice for each particle to provide updated new positions and rates of
displacement. The particle acceleration is computed and then integrated for
velocities and displacements at each time increment. DEM modeling of AC
microstructure involves the following major steps:
622 Road Materials and Pavement Design. Volume 11 – No. 3/2010

1. Input the AC microstructure


2. Define the boundary and initial conditions, and
3. Define the contact model and material properties

4.1. Mastic and aggregate particle generation


Downloaded by [Federal Highway Admin Research] at 07:39 01 August 2012

(a) (b)

(c)
Figure 2. Example to illustrate the AC model particle generation (a) mastic
particles in blue, (b) aggregate particles in red, and (c) AC PFC2D mixture model;
HL CMHB-C
Micromechanical Modeling of Asphalt Concrete Uniaxial Creep 623

The first step in the DEM simulation process is to input the AC microstructure
obtained as described above into the PFC2D input module. The built-in BALL
command generates particles and can assign mastic or aggregate attributes to them
through their Cartesian coordinate locations identified from the processed images.
An example illustrating the AC model particle generation for mixture type HL
CMHB-C is shown in Figure 2. Figures 2a and 2b show the mastic particles and the
aggregate particles, respectively. An advantage of the PFC2D is that it allows
“clumping” similar adjacent particles together using the FISH clump function (e.g.,
all elements belonging to the same aggregate can be associated together) and thus,
avoids force-displacement calculations between them. The combinations of mastic
Downloaded by [Federal Highway Admin Research] at 07:39 01 August 2012

and aggregate particles yield a realistic representation of the AC microstructure that


can be used as input to the modeling of the AC behavior (Figure 2c).

4.2. Boundary and initial conditions

Four rigid walls (platens) were defined as the boundary constraining the AC
microstructural models. The boundary walls can be moved by specifying a wall
velocity. In the case of uniaxial compression loading conditions, the upper wall can
be used to load the AC model by specifying a wall velocity, while fixing the lower
platen. A negligible confining wall velocity was assumed for the lateral walls to
simulate the actual atmospheric pressure effect on the AC core specimen inside the
testing chamber. The AC model was loaded in a strain-controlled fashion by
specifying the upper wall velocity. A numerical “servo-mechanism” implemented
via the FISH function was devised to monitor the amount of stress generated by the
top platen and alter its vertical displacement accordingly so a stress of 207 kPa is
maintained. The stresses were computed by dividing the total force acting on this
wall by the wall area. The initial velocity of the upper horizontal wall was
established through a parametric analysis with values ranging from 0.3 to 0.9 m/sec.
The reason of selecting these wall velocities was to expedite establishing creep
stress of 207 kPa. Comparisons of DEM creep predictions to test data revealed that
an initial wall velocity of 0.70 m/sec is appropriate. Similarly, the confining stresses
were kept constant by adjusting the lateral wall velocities using a numerical “servo-
mechanism” implemented through FISH functions. The specified lateral stress was
assumed equal to 0.20 MPa, a value considerably lower than the anticipated stiffness
of the AC, as recommended in the literature (Itasca Consulting Group, 2003).

4.3. Contact model and material properties

Contact models describe the force-displacement interaction between particles.


For the AC microstructure, these contacts include mastic-to-mastic, mastic-to-
aggregate, wall-to-mastic, aggregate-to-aggregate, and wall-to-aggregate. These
interactions were defined using two contact models, namely a linear contact model
624 Road Materials and Pavement Design. Volume 11 – No. 3/2010

(frictional law) and a linear viscoelastic model (i.e., the Burger model described
earlier). The current material testing practices are not sufficient to characterize all of
AC mixture interactions such as interactions of mastic-to-aggregate, aggregate-to-
aggregate, and mastic-to-mastic (non-shear). The linear elastic contact model was
used to define the aggregate-to-aggregate, wall-to-mastic, and wall-to-aggregate
contacts. Previous studies showed that an elastic modulus value of 30 GPa is
appropriate for aggregates (Abbas et al., 2005). In addition to this model, slippage
in the aggregate-to-aggregate contact was allowed by means of the contact friction
coefficient P The Burger model was used to describe the mastic-to-mastic and the
mastic-to-aggregate contacts. The Burger model constants utilized in shear direction
Downloaded by [Federal Highway Admin Research] at 07:39 01 August 2012

are given in Table 3. The elastic constants for the Burger model in the normal
direction were obtained from those in the shear direction, considering that they are
related through Equation [1] as described in the literature (e.g., Abbas, 2004)

E
2(1  Q ) [1]
G

Thus the ratio E/G is equal to the ratio of the normal contact stiffness divided by
shear contact stiffness, which is denoted in PFC2D by Kn/Ks. The stiffness of the
walls was assumed to be 10 times higher than the stiffness of the aggregate. An
additional law for the mastic-to-mastic contact was defined to allow a cohesive
breaking of the bond when the strain limits of the LVE are reached. This consistent
with the literature suggests that adhesive failure (i.e., breaking of the aggregate-to-
mastic bond commonly referred to as stripping) is not common under dry
conditions.

5. Simulation results

5.1. Sensitivity analysis

DEM requires a time step to be small enough to assure static equilibrium of


particles at all times. The PFC2D software can be used to determine automatically the
critical time step using Differential Density Scaling (DDS) procedure (using SET dt
dscale command) with a time step of unity. DDS provides a true dynamic solution
valid for both steady- and non-steady-state conditions.
The DEM simulations were carried out using a time step of 1 second for a total
of 10800 calculation cycles, to simulate the full length of the creep laboratory tests.
The results were obtained in terms of vertical strain using a virtual gauge length of
100 mm centered at the mid-height of the sample. An extensive sensitivity analysis
was conducted using aggregate-to-aggregate contact friction coefficient P and
contact stiffness ratio (Kn/Ks).
Micromechanical Modeling of Asphalt Concrete Uniaxial Creep 625

Several contact friction coefficients were selected, namely P = 0 (i.e., AC


mixture idealized as loose sand), 0.1, 0.2, 0.3, 0.4, 0.5, and f (i.e., AC mixture
idealized as rigid or intact rock). Moreover, Poisson’s ratio values of Q = 0.10, 0.20,
0.30, and 0.40 were selected yielding contact stiffness ratio of Kn/Ks = 2.2, 2.4, 2.6,
and 2.8, respectively. The contact stiffness ratio was calculated using Equation [1].
The uniaxial static creep test simulation results indicated that P significantly
affect the permanent deformation behavior of AC mixtures. It was found that Pis
inversely proportional to the slope of the axial stain steady-state region and increases
with the intercept. Moreover, the DEM predicted tertiary creep flow time increases
Downloaded by [Federal Highway Admin Research] at 07:39 01 August 2012

with POn the other hand, Kn/Ks does not appreciably affect the apparent
deformation resistance behavior of the AC model. In general, the experimentally
measured axial strain results are in good agreement with the DEM predicted values
when P= 0.5 (i.e., corresponds to an angle of internal friction of 270) and
Kn/Ks = 2.8 (i.e., corresponds to a Poisson’s ratio of 0.40) were utilized.

5.2. Axial strain versus loading time

The experimentally measured and DEM predicted total axial strains are shown in
Figures 3 to 5 for the three AC mixtures considered. In general, the shape of the
DEM predicted axial strain curves exhibit similar trends to the experimentally
measured ones. The predicted axial strain versus time responses consist of three
distinct regions, namely a primary region where the strain rate decreases, a
secondary (or steady-state) region where the strain rate is constant and a tertiary
region where the strain rate increases. The latter was not obvious in all the
experimental results analyzed in this study.
In the primary stage, the predicted axial strains were significantly lower than the
experimental strains for all AC mixture models simulated. The reason appears to be
the way PFC2D handles the build up of vertical stress to the desired level of 207 kPa.
On the other hand, in the steady-state region, the axial strains for all AC models
were in satisfactory agreement with the experimental strains. The regression
parameters of this region were used to compare the permanent deformation potential
of AC mixtures models, as presented in Table 4. The maximum absolute error
observed for the slope and the intercept of the steady-state region was lower than
2.6%. However, the experimental observations fell short to distinctly characterize
the AC model behavior in the tertiary region. The main reason is that most
laboratory creep tests were not carried out long enough for the tertiary zone to
materialize, as pointed out earlier.
626 Road Materials and Pavement Design. Volume 11 – No. 3/2010

Table 4. Axial strain steady-state region regression parameters

Intercept Absolute Slope Absolute


Mixture (mm/mm) x10-6 Error (mm/mm)/sec x10-7 Error
ID (%) (%)
Measured Predicted Measured Predicted

HL CMHB-C 3820 3900 2.09 2.62 2.64 0.76

G CMHB-C 4160 4150 0.24 3.83 3.85 0.52


Downloaded by [Federal Highway Admin Research] at 07:39 01 August 2012

SL CMHB-C 4240 4350 2.59 5.80 5.70 1.72

HL: Hard Limestone; G: Granite; SL: Soft Limestone.

12000

10000
Total Axial Strain (10 ) (mm/mm)

8000
-6

6000

4000 DEM Predicted


Expt. Measured
2000

0
0 2000 4000 6000 8000 10000 12000 14000
Time (seconds)

Figure 3. Measured versus predicted axial strain; HL CMHB-C

12000

10000
Total Axial Strain (10 ) (mm/mm)

8000
-6

6000
DEM Predicted

4000 Expt. Measured

2000

0
0 2000 4000 6000 8000 10000 12000 14000
Time (seconds)

Figure 4. Measured versus predicted axial strain; G CMHB-C


Micromechanical Modeling of Asphalt Concrete Uniaxial Creep 627

12000

10000
Total Axial Strain (10 ) (mm/mm)

8000
-6

Expt. Measured
6000

DEM Predicted
4000
Downloaded by [Federal Highway Admin Research] at 07:39 01 August 2012

2000

0
0 2000 4000 6000 8000 10000 12000 14000
Time (seconds)

Figure 5. Measured versus predicted axial strain; SL CMHB-C

5.3. Creep compliance versus log loading time

Table 5. Creep compliance steady-state region regression parameters

Intercept Slope Absolute


Mixture (1/kPa) x10-6 Absolute (1/kPa.sec) x10-7 Error
ID Error (%)
(%)
Measured Predicted Measured Predicted

HL
0.85 0.84 1.17 3.01 3.05 1.33
CMHB-C

G
1.00 0.98 2.00 4.19 4.17 0.48
CMHB-C

SL
0.80 0.78 2.50 4.51 4.53 0.44
CMHB-C

HL: Hard Limestone; G: Granite; SL: Soft Limestone.

The main advantage of plotting the creep compliance (i.e., the inverse of
stiffness) versus the logarithm of the loading time is that it allows to easily identify
the three distinct creep stages described earlier. The creep compliance was computed
by dividing the axial strain by the applied time-dependent stress. The experimentally
measured and DEM predicted creep compliance versus logarithm time curves are
shown in Figures 6 to 8. These figures confirm that the DEM simulation
underestimates the creep compliance in the initial stages of loading, while do a
satisfactory predictions in the steady-state creep stage. Table 5 shows a summary of
628 Road Materials and Pavement Design. Volume 11 – No. 3/2010

the regression parameters of this region and a comparison to similar values obtained
from the experimental data. Again, the maximum absolute error observed for the
slope and intercept of steady-state region was lower than 2.5%. No meaningful
comparison of the flow time values obtained by the models could be made with
those obtained from the experimental data, as explained earlier.

-4
10
Log Axial Creep Compliance (1/kPa)
Downloaded by [Federal Highway Admin Research] at 07:39 01 August 2012

-5
10

Expt. Measured

-6
10

-7
DEM Predicted
10

-8
10 -1 0 1 2 3 4 5
10 10 10 10 10 10 10
Log Time (seconds)

Figure 6. Measured versus predicted creep compliance; HL CMHB-C

-4
10
Log Axial Creep Compliance (1/kPa)

-5
10
Expt. Measured

-6
10

-7 DEM Predicted
10

-8
10 -1 0 1 2 3 4 5
10 10 10 10 10 10 10
Log Time (seconds)

Figure 7. Measured versus predicted creep compliance; G CMHB-C


Micromechanical Modeling of Asphalt Concrete Uniaxial Creep 629

-4
10
Log Axial Creep Compliance (1/kPa)

-5
10

Expt. Measured
-6
10

-7
10
DEM Predicted
Downloaded by [Federal Highway Admin Research] at 07:39 01 August 2012

-8
10

-9
10 -1 0 1 2 3 4 5
10 10 10 10 10 10 10
Log Time (seconds)

Figure 8. Measured versus predicted creep compliance; SL CMHB-C

6. Discussions

The primary object of the paper was to present a DEM simulation of the creep
behavior of AC mixtures and compare to experimental creep data obtained under
uniaxial loading. For this purpose, a DEM software package PFC2D was utilized. A
high-resolution X-Ray CT facility was utilized to capture the AC microstructure.
State-of-the-art DIP technique, VGM, was utilized to process the AC X-Ray CT
images in a format suitable for input into the PFC2D simulation. Rheological tests on
mastics were performed using DSR device to characterize their viscoelastic
properties. Two oscillatory tests were carried out, namely amplitude sweep test and
frequency sweep tests. The former was used to determine the LVE range and the
maximum limiting shear strain parameter. The Burger model parameters were
characterized by utilizing the later test results.
The DEM modeling approach utilized had a number of limitations highlighted
below:
– The DEM approach was two-dimensional. Therefore any three-dimensional
effects of the AC microstructure were lost.
– The maximum image resolution obtained was 195µm. Hence, all aggregate
particles smaller than about Sieve No. 70 could not be identified.
– The binder viscoelastic behavior was characterized in shear only due to lack of
a suitable high-temperature test that allows binder characterization under normal
stresses. As a result, assuming that the shear and normal viscoelastic properties were
related through the Poisson’s ratio was unavoidable. Furthermore, this value of the
Poisson’s ratio had to be assumed.
630 Road Materials and Pavement Design. Volume 11 – No. 3/2010

– The handling of the upper boundary platen by PFC2D generated solution


instability in the initial stages of creep load imposition.
– Negligible confining lateral pressure was assumed. However, the AC mixtures
were tested under unconfined compression tests.
– The cohesive failure assumed for the mastic was shear strain-level based. In
reality, cohesive failure takes place in response to limiting shear and normal
stress/strain.
Considering the above mentioned research challenges, the future scope of the
study include the following:
Downloaded by [Federal Highway Admin Research] at 07:39 01 August 2012

– VGM can be easily modified to process the AC X-Ray CT images for three-
dimensional analysis. Thus, realistic representation of the AC mixture model can be
obtained by utilizing PFC3D software. Thus, the in-situ or laboratory tested material
responses can be truly simulated to predict intrinsic micromechanical behavior of
the AC mixtures.
– The Burger model needs to be generalized (i.e., include a larger number of
viscous and damping elements) to better describe the complex mastic behavior.
– The mastic failure criteria need to be generalized to account for the detailed
normal and shear stress/strain state at the contacts.
– Failure criteria need to be expanded to describe adhesive failure (i.e., breaking
of the bond between aggregate and mastic) which is typical under high moisture
conditions.
– The need to implement the last two failure criteria will require the development
of user-defined visco-elasto-plastic contact models into DEM software.

7. Summary and conclusions

This paper presented a DEM approach for modeling the creep behavior of
asphalt concretes. Their microstructure was captured through X-Ray CT techniques
and automatically processed as input to the PFC2D software. One mix design and one
binder grade were used combined with aggregates of similar gradation but from
three different sources, namely Hard Limestone (HL), Soft Limestone (SL) and
Granite (G). The Burger model was fitted to describe the viscoelastic properties of
the mastics using actual DSR measurements. The simulation results were compared
to uniaxial creep test data obtained on the same AC cores at testing temperature of
60°C. The models underestimated the creep compliance of the mixtures in the
primary creep stage due to the way the vertical load application was simulated. On
the other hand, they predicted quite accurately creep compliance in the secondary
creep compression stage. The maximum absolute error observed for the slope and
intercept of this region was not higher than 2.6%. Although the models predicted
flow time, no meaningful comparison could be made with experimental flow times.
Micromechanical Modeling of Asphalt Concrete Uniaxial Creep 631

Acknowledgment
Sincere thanks are expressed to Dr. Nazarian of the University of Texas-El Paso
and Dr. Masad of Texas A&M University for supplying the AC laboratory test
results and the asphalt concrete X-Ray CT images, respectively.

8. References

Abbas A., Simulation of the micromechanical behavior of asphalt mixtures using the discrete
Downloaded by [Federal Highway Admin Research] at 07:39 01 August 2012

element method, Ph.D. Dissertation, Civil Engineering Department, Washington State


University, Pullman, WA, 2004.
Abbas A., Papagiannakis A.T., Masad E., Shenoy A., “Modeling asphalt mastic stiffness
using discrete element analysis and micromechanics-based models”, The International
Journal of Pavement Engineering, Vol. 6, No. 2, 2005, p. 137-146.
Alvarado C., Mahmoud E., Abdallah I., Masad E., Nazarian S., Langford R., Tandon V.,
Button J., “Feasibility of quantifying the role of coarse aggregate strength on resistance to
load in HMA”, TxDOT Project No. 0-5268 and Research Report No. 0-5268, 2007.
Baumgaertel M., Winter H.H., “Determination of discrete relaxation and retardation time
spectra from dynamic mechanical data”, Rheological Acta, Vol. 28, 1989, p. 511-519.
Buttlar W.G., You Z., “Discrete element modeling of asphalt concrete: a micro-fabric
approach”, Transportation Research Record 1757, Transportation Research Board,
National Research Council, Washington, D.C., 2001, p. 111-118.
Chang K.G., Meegoda J.N., “Micromechanical simulation of hot mix asphalt”, Journal of
Engineering Mechanics, Vol. 123, No. 5, 1997, p. 495-503.
Collop A.C., McDowell G.R., Lee Y.W., “Modeling dilation in an idealized asphalt mixture
using discrete element modeling”, Granular Matter, Vol. 8, 2006, p. 175-184.
Cundall P A., “A computer model for simulating progressive large scale movements in blocky
rock systems”, Proc., Symp. of the Int. Society of Rock Mechanics, Nancy, France, Vol. 1,
1971, Paper No. II-8.
Cundall P.A., Strack O.D., “Discrete numerical model for granular assemblies”,
Geotechnique, Vol. 29, No. 1, 1979, p. 47-65.
Dai Q., You Z., “Prediction of creep stiffness of asphalt mixture with micromechanical finite
element and discrete element models”, Journal of Engineering Mechanics, Vol. 133,
No. 2, 2007, p. 163-173.
Itasca Consulting Group, “Particle flow code in two-dimensions (PFC2D) manual version 3.1”,
Itasca Consulting Group, MN, 2003.
Kim Y., Little D., “Linear viscoelastic analysis of asphalt mastics”, Journal of Materials in
Civil Engineering, Vol. 16, No. 2, 2004, p. 122-132.
Masad E., Jandhyala V.K., Dasgupta N., Somadevan N., Sashidhar N., “Characterization of
air voids distribution in asphalt mixes using x-Ray computed tomography”, Journal of
Materials in Civil Engineering, Vol. 14, No. 2, 2002, p. 122-129.
632 Road Materials and Pavement Design. Volume 11 – No. 3/2010

Masad E., Muhunthan B., Shashidhar N., Harman T., “Internal structure characterization of
asphalt concrete using image analysis”, Journal of Computing in Civil Engineering, Vol.
13, No. 2, 1999, p. 88-95.
Papagiannakis A.T., Abbas A., Masad E., “Micromechanical analysis of viscoelastic
properties of asphalt concrete”, Transportation Research Record 1789, Transportation
Research Board, National Research Council, Washington, D.C., 2002, p. 113-120.
Rothenburg L., Bogobowicz A., Haas R., Jung F.W., Kennepohl G., “Micromechanical
modelling of asphalt concrete in connection with pavement rutting problems”,
Proceedings of the 7th International Conference on Asphalt Pavements, 1992, p. 230-245.
Downloaded by [Federal Highway Admin Research] at 07:39 01 August 2012

Shashidhar N., Zhong X., Shenoy A.V., Bastian E.J., “Investigating the role of aggregate
structure in asphalt pavements”, 8th Annual Symposium Proceedings on Aggregates,
Asphalt Concrete, Base, and Fines, Denver Co., 2000.
Shashidhar N., “X-ray tomography of asphalt concrete”, Transportation Research Record
1681, Transportation Research Board, National Research Council, Washington, D.C.,
1999, p. 186-192.
Sousa J.B., Craus J., Monismith C.L., “Summary report on permanent deformation in asphalt
concrete”, Strategic Highway Research Program, National Research Council,
Washington, D.C., Report No. SHRTP-A/IR-91-104, 1991.
Tashman L., Masad E., Angelo J.D., Bukowski J., and Harman T., “X-Ray tomography to
characterize air void distribution in Superpave gyratory compacted specimens”,
International Journal of Pavement Engineering, Vol. 3, No. 1, 2002, p. 19-28.
You Z., Buttlar W.G., “Discrete element modeling to predict the modulus of asphalt concrete
mixtures”, Journal of Materials in Civil Engineering, Vol. 16, No. 2, 2004, p. 140-146.
You Z Buttlar W.G., “Stiffness prediction of hot mixture asphalt (HMA) based upon
microfabric discrete element modeling (MDEM)”, Proceeding of 4th International
Conference on Road and Airfield Pavement Technology, China, Vol. 1, 2002, p. 409-417.
Yu L., Shi-Feng R., Guang-Hu X., “Discrete Element Simulation of Asphalt Mastics Based
on Burgers Model”, Journal of Southwest Jiaotong University Journal of Southwest
Butler University, Vol. 1, No. 15,2007, p. 20-26.
Zelelew H.M., Papagiannakis A.T., “A Volumetrics Thresholding Algorithm for Processing
Asphalt Concrete X-Ray CT Images”, International Journal of Pavement Engineering,
2008a (In-press).
Zelelew H.M., Papagiannakis A.T., “A Wavelet-based Characterization of Aggregate
Segregation in Asphalt Concrete X-Ray Images”, International Journal of Pavement
Engineering, 2008b (In-press).
Zelelew H.M., Papagiannakis A.T., “Effects of the Rheological Properties of Mastics on
Creep performance of Asphalt concretes”, International Conference on Transport
Infrastructures (ICTI), São Paulo, Brazil, 2010 (In-review).

Received: 20 October 2008


Accepted: 12 April 2010

Potrebbero piacerti anche