Hindawi

Advances in Civil Engineering

Volume 2018, Article ID 5180562, 11 pages
https://doi.org/10.1155/2018/5180562

Research Article
Pavement Analysis and Design by Multiphysics Reconstructing
Algorithm for the Virtual Asphalt Mixture Based on the
Discrete-Element Method

Danhua Wang,1 Xunhao Ding ,2 Tao Ma ,2 Weiguang Zhang,2 and Deyu Zhang3
1
School of Computer Engineering, Nanjing Institute of Technology, Nanjing, Jiangsu 211167, China
2
School of Transportation, Southeast University, 2 Sipailou, Nanjing, Jiangsu 210096, China
3
School of Engineering and Architecture, Nanjing Institute of Technology, Nanjing, Jiangsu 211167, China

Correspondence should be addressed to Tao Ma; [email protected]

Received 26 February 2018; Accepted 8 April 2018; Published 7 May 2018

Academic Editor: Fan Gu

Copyright © 2018 Danhua Wang et al. This is an open access article distributed under the Creative Commons Attribution License,
which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

Based on the Particle Flow Code in Two dimensions (PFC2D), an algorithm for modeling the two-dimensional virtual asphalt
mixture was proposed in this study. By combining the AIMS scanning technology (Aggregate Imaging Measurement System) with
the designed stochastic algorithm, the virtual coarse aggregates could be generated rapidly and precisely. Different from the
conventional methods, the contour shapes of the coarse aggregates were rebuilt only to balance the shape modeling precision and
simulation efficiency. Then by distributing the coarse particles within container, virtual skeletons were formed firstly. An innovate
algorithm was proposed afterwards to distinguish the external and internal area of the coarse aggregates and then model the
mastic part by filling the irregular hollow shape with uniformly arranged balls. By deleting the mastic balls randomly, the voids
were reconstructed consistent with the actual ratio. In the end, the virtual uniaxial compressive tests of AC-16 were simulated
within PFC2D and the dynamic modulus at different load frequencies was predicted. The results indicated that the proposed
algorithm could not only model the asphalt mixture precisely but also characterized its mechanical behavior as well.

1. Introduction by Airey et al. [11] to reveal the inner mechanism. In their

tests, two asphalt mixtures including a continuous and
Asphalt mixtures are three-phase structures consisting of gap-graded gradation were compacted and compared. The
aggregates, asphalt binder, and air voids. It has been results indicated that gap-graded asphalt mixture was de-
highlighted by many researchers that the mechanical per- graded more easily compared to the continuously grade
formance of asphalt mixture was influenced by its hetero- mixture. Isailović et al. [12] pointed out that asphalt ageing
geneous components significantly [1–7]. Tests conducted by played an important role in reducing the effects of material
Liu and Dai [8] focused on the gradation influence on the rut recovery based on fatigue tests. In addition, the mixtures
resistance of the asphalt mixtures. The results showed that prepared with additionally 0.5% of bitumen by mass showed
based on the Superpave Gyratory Compactors (SGC), the the best recovery characteristics compared to others. The
optimum asphalt content is lower 0.2%∼0.5% than the one void effects on fatigue life of asphalt mixture were analyzed
by the Marshall method. As the aggregate size increased, the by Hasan and Ahmad in early 1973 [13]. The voids within
mixtures could have larger resistance compared to the mixtures were varied by the changes of asphalt content and
smaller particles. Similar results were also concluded based aggregate gradation. The fatigue response of different mixtures
on the studies of Coleri et al. [9] and Gokhale et al., re- was investigated showing the important roles the voids played
spectively [10]. Focusing on the aggregate degradation in. Apart from the fatigue performance, the rutting-resistance,
during compaction, several designed tests were carried out moisture damage-resistance, and strength of asphalt mixture
2 Advances in Civil Engineering

would also be influenced by air void, which was founded by modeling which can be generated in quantities rapidly and
some researchers [14, 15]. Current design methods for asphalt maintain the realistic shapes at the same time.
mixtures mostly focused on the volumetric properties (ag-
gregate gradation, asphalt content, and void) and paid little
attention on the morphological characteristics of the com- 2. Objective and Scope
ponents which were really hard to investigate just by labo-
ratory apparatus. It is believed that, apart from the volumetric The objective of this study is to rebuild the virtual asphalt
properties, the shapes of the inner components, especially the mixtures rapidly and precisely based on the Discrete-
aggregates, indeed have impact on the properties of asphalt Element Method (DEM). To achieve this, with the help of
mixtures as well [16–18]. AIMS scanning technology, an algorithm was designed for
As an effective numerical technology, the Discrete- modeling coarse aggregates, asphalt mastic, and voids. The
Element Method (DEM) was proposed by Cundall [19], relevant issues include the shape and size measurements
and then a software named as Particle Flow Code (PFC) was through AIMS, irregular particle reconstitution derived
developed to help the mechanical calculation of heteroge- from the standard virtual aggregates, asphalt mastic gen-
neous materials, which has been widely used in many fields eration by irregular area judgments, random voids model-
[20, 21]. Not only the macromechanical behavior but also the ing, and the mechanical behavior prediction.
microresponse of the asphalt mixtures could be predicted
based on the PFC, and many virtual tests were conducted
3. Shape Measurement through Scanning
under complex conditions for various goals by some re-
searchers [22–24]. However, prior to simulations, it is sig- The AIMS apparatus was utilized to record the particle
nificantly important to develop the precise virtual models shapes including the angularity index and size index. While
based on their volumetric and shape property which is scanning the samples, coarse aggregates should be put into
a guarantee for the results validations [25]. Thus, many slots of designed trays. With the trays rotated slowly, each
methods were proposed for heterogeneous component re- particle would go through the scanning area, and a digital
constitution within asphalt mixtures which fell into two camera was set to capture their features. Shape properties of
categories in general. One is the random model and another particles were calculated and stored and could be output for
is the image-based models. The random models tended to the virtual modeling within PFC2D. The related shape
develop the virtual particles by some designed stochastic measurements are introduced as follows.
algorithm. By changing the control parameters related to
shapes and sizes, a large number of the simplified virtual
shapes could be developed rapidly. Such methods were
3.1. Angularity Index. Angularity index applies to coarse
proposed by Lu and Mcdowell [25], Das et al. [26], and
aggregate sizes and describes variations at the particle
Zhang et al. [27]. In Lu and Mcdowell’s studies, the virtual
boundary that influence the overall shape. The angularity
particles were modeled by the overlapping balls [25], and the
index quantifies changes along a particle boundary with
modeling algorithm was optimized further by Das et al. [26].
higher gradient values indicating a more angular shape.
Coarse aggregates were assumed to be hexahedrons, pen-
Angularity index has a relative scale of 0 to 10000 with
tahedrons, and tetrahedrons by Zhang et al. [27]. By cutting
a perfect circle having a small nonzero value as shown in
the particle clumps randomly with the help of three vari-
Figure 1. The angularity index is analyzed by quantifying the
ables, the irregular shape was rebuilt preliminarily which was
change in the gradient on a particle boundary and is related
roughly consistent with the realistic ones. The random
to the sharpness of the corners of 2-dimensional images of
models for virtual particles could be generated in larger
aggregate particles as shown in the following equation:
amounts quickly without preparing specimens in laboratory.
n−3􏼌 􏼌􏼌
Although it was performed with better efficiency compared 1 􏼌
to the image-based models, the shape modeling was not GA � 􏽘 􏼌􏼌θi − θi+3 􏼌􏼌, (1)
(n/3) − 1 i�1
precise enough with excessive simplifications. The image-
based models are mostly generated with the help of the X-ray where θ is the angle of orientation of the edge points; n is the
computed tomography (CT) scanning technology [28]. By total number of points; and subscript i denotes the ith point
processing the component images from the scanning, the on the edge of the particle.
heterogeneous materials could be reconstructed within PFC
precisely. Such methods were introduced and utilized in
some studies [29–33]. However, the use of the image-based 3.2. Size Index. The particle’s shortest, intermediate, and
models are limited by the test environments and conditions longest lengths are measured to describe the size charac-
which is a barrier preventing the virtual modeling to be teristics as shown in Figure 2. The value L is the particle’s
conducted efficiently. First, a scanning device is must which longest length representing the largest size in all the di-
provides the fundamental images. Second, when rebuilding rections. And the particle’s intermediate length w is the
image-based models, test specimens should be developed for largest size in the planes perpendicular to the particle’s
scanning in advance. Therefore, it is time-consumed and longest length. The particle’s shortest length t is the largest
cannot meet the needs of large amount simulation tasks. size of the planes perpendicular to the particle’s longest and
Thus, further studies are needed to optimize the particle intermediate lengths at the same time.
Advances in Civil Engineering 3

(a) (b) (c) (d)

Figure 1: Angularity index for various shapes. (a) 0 ≤ low ≤ 2100; (b) 2100 < moderate ≤ 3975; (c) 3975 < high ≤ 5400; (d) 5400 < extreme ≤ 10000.

w
t

(a) (b)

Figure 2: Size index for aggregate particles. (a) Side view; (b) top
view.

4. Algorithm for Developing Virtual (a) (b)

Coarse Aggregates Figure 3: Rebuilding virtual particles by filling contours with balls.
4.1. Standard Particles Generations. The virtual coarse ag-
gregates were rebuilt within PFC2D as shown in Figure 3. methods. More angular shapes could also be modeled in the
It is noted that with the goal of developing virtual spec- same way depending on the simulation needs but were not
imen rapidly and randomly, not all the particles in actual included here. Since it is very time-consumed to rebuild
mixtures should be scanned which would consume ex- each particle in reality, the representative shape modeling,
cessive time. It is meaningful to conduct simulations by as a more efficient mean, could save time and develop
generating representative virtual particles with enough virtual mixtures rapidly. In this study, the following pro-
irregular shapes. Thus, as an effective mean, when con- posed algorithm was introduced emphatically.
ducting other simulations, the representative virtual Assumptions were made that there was no fracture or
particles can be generated rapidly without scanning the break of virtual particles; thus, the inner balls of the virtual
components again. particles could be released which could conduct the simu-
As shown in (1), it is known that the angularity index lations more effectively. The particle surfaces would bear the
quantifies changes along a particle boundary regardless of contact force and were simulated by the contour balls as
its size. Thus, shapes were modeled firstly without size shown in Figure 3. Compared to the solid virtual particles,
variations. Enough coarse aggregates were selected firstly the mass of the proposed methods was smaller than the
for rebuilding the standard particles with their inter- reality and should be calibrated for all virtual particles as in
mediate lengths w (defined as the minimum sieve size that the following equation.
the particle is able to pass) converted to 19 mm compul-
m × π × r2 m ×(rk)2
sively. And the areas of all the adjusted particles were p2  p1 ×  p 1 × 4 × , (2)
measured. Then, the particles of other sizes were derived n × π ×(0.5/k)2 n
from the standard ones by size adjustments. While gen- where p2 is the calibrated density (g/cm3), p1 is the initial
erating the standard particles, the selected samples for density (g/cm3), m is the number of retained pixels, n is the
particle reconstitution should have a large angularity index number of all the pixels in Figure 1(a), k is the scaling ratio in
range from 1000 to 6500 which most realistic aggregates image processing, and r is the radius of the ball located in the
are distributed in. Then, the shape images were processed contour (mm).
by filtering the redundant black pixels. Only some key
black pixels in the shape contour were retained and then
were imported to the PFC2D. The key black pixels could 4.2. Particle Size Adjustments. Each standard particle was
describe the boundary changes and capture the main coded and saved as an executable file. When generations
properties of the irregular shapes. By generating balls in the started, corresponding code files were evoked and executed
positions of the retained pixels, the virtual particles could by PFC2D. To model the size variations, a designed controls
be developed as a rigid body through the input Fish parameter known as Sr were embedded in the code files, as
command “clump,” as shown in Figure 3. 27 particles with shown in Table 1. As shown, the size ratio is defined as the
various angular shapes were rebuilt by the proposed intermediate length ratio between the other grade aggregates
4 Advances in Civil Engineering

Table 1: Size adjustments for simulations.

Size (mm) Size ratio Controls parameter Size (mm) Size ratio Controls parameter
13.2 16 13.2 16 13.2
19 1 1 13.2–16  ,  + −  × urand
19 19 19 19 19
25 25 9.5 13.2 9.5 13.2 9.5
19–25 1,  1+ − 1 × urand 9.5–13.2  ,  + −  × urand
19 19 19 19 19 19 19
16 16 16 4.75 9.5 4.75 9.5 4.75
16–19  , 1 + 1 −  × urand 4.75–9.5  ,  + −  × urand
19 19 19 19 19 19 19 19

(1) Determine the size of the virtual container and

4.75 mm develop corresponding walls to model the boundary
conditions.
9.5 mm
(2) Start the coarse particles generation with the size
ranging from 16 mm to 19 mm. The virtual particles
13.2 mm are generated one by one and the shape is selected
randomly by executing the particle code files.
(3) Develop three variables known as x1 , y1 , and rot
16 mm which is calculated as shown in (3). These three
variables are all drawn from the uniform distribution
with different ranges. Before a virtual particle is
19 mm
generated, x1 , y1 , and rot embedded in the code files
Figure 4: Virtual size reconstitutions based on standard particles. are determined randomly. And then the local co-
ordinate system (coordinate system of particles) is
and the standard ones. And the size ratios of each grade converted to a global coordinate system (coordinate
aggregates were coded with the help of the input “urand” system of container). By using the random values of
value which was drawn from the uniform distribution U(0, 1 x1 , y1 , and rot, the coarse particles can be generated
within PFC2D. Before the code files were evoked, the Sr was at a random position within the container area.
determined randomly by the system leading to the size
generations. Moreover, the areas of different grades were x1  U(0, xw ),
adjusted together by multiplying the standard particle area y1  U 0, yh , (3)
with their size ratio squared.
Figure 4 shows the virtual size reconstitutions based on rot  U(0, 2π),
standard particles. As shown, the standard particles with
where x1 and y1 are the random x- and y-coordinates of the
19 mm intermediate lengths were developed firstly, and then
initial positions, respectively; rot is the random angle of the
the others were derived from the standard ones with same
generated particles. xw and yh are the width and height of
angular shapes. By this way, 135 virtual particles with 27
the virtual containers.
angular shapes of 5 different grades were developed rapidly.
(4) Check the overlap of the generated particles. Since
the particles are generated within container area one
4.3. Coarse Particles Distribution within Mixture. The virtual by one. So when a particle has been generated at
asphalt mixtures were developed in 3 steps: the coarse ag- a random position, it is necessary to check if it is
gregate generation, the asphalt mastic filling, and the voids overlapped with the others. If there is no overlap, the
modeling. It is unpractical to rebuild all the particles’ shapes particle can be generated, else, change x1 , y1 , and rot
especially the numerous finer aggregates, which will exceed until the particle can be generated correctly;
the computer capacity limitations. On the other side, the
(5) Start to generate the next particle by cycling from
shape influences of the finer aggregates are not obvious to
step 2 to 4 until reach the required mapping area of
some extent compared to the coarse parts. So the major
16–19 mm particles.
assumptions are made that the aggregates finer than
4.75 mm are regarded as the part of the asphalt mastic and By the steps from 1 to 5, the virtual particles with the size
are modeled by round balls directly within PFC2D. ranging from 16 mm to 19 mm could be developed suc-
Coarse aggregates were distributed randomly within the cessfully. Then similar processes were carried out for the
container area. The gradations of particles and the asphalt particles of other grades as shown in Figure 5. To distribute
content were calculated and converted to mapping area in the different grades particles correctly, the particles with
two dimensions. The mapping area of different grades largest sizes should be generated first and then the smaller
particles were taken as control values which determined the ones. For example, start the particle generation of 16–19 mm
end time of coarse aggregate generations. The coarse ag- firstly, and then followed by 13.2–16 mm, 9.5–13.2 mm, and
gregates distributions were conducted as follows: 4.75–9.5 mm in turns.
Advances in Civil Engineering 5

Control line

Nearest ball in
right side

Point A

Virtual wall

Figure 6: Distinguishing the position of mastic balls between the

particle clump and the wall on the left side.

Virtual wall

Point A

Figure 5: Coarse particles distributions within the virtual container.

5. Algorithm for Developing Asphalt Mixture Control line

5.1. Asphalt Mastic Modeling. The asphalt mastic was

modeled by the uniform arranged balls with a diameter of Nearest ball in
0.001 m. As common practices, the conventional methods left side
tended to fill the particle external area with finer balls (mastic
part) by judging if there are balls already [27]. Because the Figure 7: Distinguishing the position of mastic balls between the
virtual particles in conventional methods were filled with particle clump and the wall on the right side.
balls (solid shape), the external and internal areas of the
particle could be distinguished easily by balls judgments.
However, it is hard to fill in the mastic balls directly when it
comes to the coarse aggregate skeleton developed in this
study, as shown in Figure 5. The differences were not obvious Point B
between the external and internal particle areas which were
all irregular in shape. To fill the irregular hollow shape
(particle external area) with balls and ignore the particle
internal area, the algorithm for the asphalt mastic was in-
troduced as followed.
Control line Point A
(1) Measure the size of total specimens to get all the
coordinate values which are prepared for filling
judgments.
(2) Develop two parameters known as kright and kleft to Figure 8: Distinguishing the position of mastic balls between two
record the position properties. They are all set zero particle clumps (inside or outside the aggregates).
initially. Then check all the coordinate values from
left to right and bottom to top in turns. The positions
(1) If there is a nearest particle ball in the right side,
will fall into four categories as shown in Figures 6–9.
convert kright from 0 to 1, else, kright is still equal
Before a group of coordinate values is checked, two
to 0.
parallel control lines are utilized with a 0.001 m
(2) If there is a nearest particle ball in left side,
interval (diameter of the mastic ball). Check the two
convert kleft from 0 to 1, else, kleft is still equal
sides of the position and find the nearest entities
to 0.
(walls and particle balls without mastic balls gen-
erated already) between two control lines as follows. Then the external and internal area of the particles can be
The control lines were not real lines plotted in the distinguished by the combination of kright and kleft . If kright
images but a designed routine coded in MATLAB to and kleft are equal to 1 and 0, respectively, the position is
help the distance judgments. between the particle clump and the left wall as shown in
6 Advances in Civil Engineering

Point A
Control line

Figure 9: Distinguishing the position of mastic balls without any

entities nearby in two sides.

Figure 6. Then the coordinate value is effective, and a mastic Figure 10: Filling the irregular area with balls to model the asphalt
ball with the diameter of 0.001 m is generated here. mastic.
If kright and kleft are equal to 0 and 1, respectively, the
position is between the particle clump and the right wall as
shown in Figure 7. Then the coordinate value is effective, and The balls without any contact force during the short-time
a mastic ball with the diameter of 0.001 m is generated here simulation were identified as the escaped balls and should be
as well. deleted as shown in Figure 11.
If kright and kleft are equal to 1 and 1, respectively, two The void content in three dimensions was converted to
conditions should be analyzed, as shown in Figure 8. As mapping area in two dimensions which was the same with
shown, the points A and B are between two particle clumps the coarse aggregates and asphalt mastic. Since the diameters
with same kright and kleft . The point A is inside the aggregates of the mastic balls have been known, the number of the void
while B is outside the model between two different particles. balls could be determined. By deleting the mastic balls
In this condition, the nearest particle balls on both sides were randomly until the number of the void balls reached re-
analyzed further. With the help of the command pointer, the quirements, the virtual void could be modeled well as shown
clump id of all the balls can be read and recognized. So in Figure 12(d).
comparing the clump id of the nearest balls, it is easy to
determine whether the nearest balls belong to a same clump.
If the clump id of two balls is equal, they come from the same 5.3. Example of Generating a Virtual Specimen. The asphalt
particle clump. Otherwise, they belong to a different particle mixture of AC-16 was developed for simulation. The gra-
clump. Thus, if the nearest balls on left and right sides belong dation of the AC-16 is shown in Table 2. Based on the
to a same particle clump, as the point A shown in Figure 8, gradation, the mapping areas of three-phase structures in-
the coordinate value is noneffective without any mastic balls cluding the coarse aggregates, asphalt mastic, and the void
generated here. When the nearest balls belong to different content were calculated using (4) and (5). And the results of
clumps, a mastic ball should be generated. the AC-16 were summarized in Table 3.
A specific condition should also be included as shown in π × D2
Figure 9 due to the space between coarse aggregates. kright M�ρ× × h, (4)
and kleft are 0 and 0, respectively, when the space is enough 4
among particles. When comes to this condition, mastic balls Where M is the total mass of the specimen in three di-
should also be generated. Following the proposed rules and mensions, g; ρ is the density of the specimen, g/cm3; D is the
coded it within PFC2D, the virtual asphalt mastic could be diameter of the specimen, cm; and h is the height of the
rebuilt successfully as shown in Figure 10. As shown, the specimen, cm.
uniform arranged balls were filled in the external particle When it comes to the two-dimensional specimen, the
area to model the mastic while the coarse aggregates kept mass can be converted as following:
same shapes and positions all the time. M V×ρ
m� ×S� × D × h � Dhρ, (5)
V V
5.2. Escaped Balls Deletion and Voids Modeling. Prior to where m is the converted mass of specimen in two di-
modeling the voids, enough calculation steps should be done mensions (g), V is the total volume of the specimen in three
within PFC2D to make the ball system stable. The mastic dimensions (cm3), S is the area of the specimen in two
balls would move continuously until reach the equilibrium dimensions (cm2), and the others are the same as those of
state. However, due to the initial overlaps among balls three dimensions.
(Figure 10), some of the mastic balls would go through the The void content was 4% and the asphalt content was
particle boundary and then escaped. So after the ball system 4.5% in mixtures. Assumptions were made that the air void
reached the equilibrium state, an initial upward velocity was is distributed uniformly in mixtures so the void contents in
assigned to all balls and then simulated for a very short time. two dimensions and three dimensions can be set as a same
Advances in Civil Engineering 7

Figure 11: Escaped balls deletion before random voids modelling.

(a) (b) (c)

(d) (e)

Figure 12: Modeling process of virtual asphalt mixtures: (a) aggregate skeleton; (b) uniform balls filling; (c) virtual asphalt mixture;
(d) voids modeling; (e) asphalt mastic modelling.
8 Advances in Civil Engineering

Table 2: Gradations for AC-16.

Gradation Passing ratio (%)for different sieving size (mm)
19 16 13.2 9.5 4.75 2.36 1.18 0.6 0.3 0.15 0.075
AC-16 100 95 84 70 48 34 24.5 17.5 12.5 9.5 6.0

Table 3: Mapping area calculations for three-phase structures. Table 6: Microparameters of selected constitutive models for
particles and walls.
Total mapping Total mapping Total mapping
Container
area for void area for coarse area for asphalt Normal stiffness, Shear stiffness, Friction
area (m2)
(m2) aggregates (m2) mastic (m2) kn (N/m) ks (N/m) coefficient
0.015 0.0006 0.0071 0.0073 4.75 2.2e6 1.5e6 0.6
9.5 2.5e6 2.5e6 0.55
13.2 1.2e6 1.2e6 0.5
Table 4: Summaries of the coarse aggregates. 16 1.8e6 1.8e6 0.4
Wall 1e10 1e10 0.35
Size (mm) 16 13.2 9.5 4.75
Required mapping
6.83e − 4 1.50e − 3 1.91e − 3 3.01e − 3
area (m2) mixtures were formed in a cylinder container with a height
Generated area (m2) 7.18e − 4 1.57e − 3 1.95e − 3 3.06e − 3 of 150 mm and a width of 100 mm. A Superpave gyratory
Error (%) 5.1 4.7 2.1 1.7 compactor was utilized to compact the asphalt mixture to
a targeted void level of 4% at the 4.5% asphalt content. The
Table 5: Summaries of the asphalt mastic and voids. specimens were tested at 0°C and 138 kPa confining pressure
and the loading frequencies are 0.1, 1, 5, 10, and 25 Hz,
A B C D E F Error (%) respectively.
7.3e − 3 7.301e − 3 9300 10076 764 12 0.014
A: required mapping area for asphalt mastic (m2); B: generated mapping area
for asphalt mastic (m2); C: required number of balls within asphalt mastic; D: 6.2. Simulations for Dynamic Modulus Prediction. The virtual
actual number of balls within asphalt mastic before void modeling; E: number asphalt mixtures were developed as shown in Figure 12(c).
of the deleted balls for modeling voids; F: number of the deleted balls which
have been escaped.
When simulated, the top and bottom walls were modeled as
a sine load while the walls on left and right sides kept
a constant confining stress based on the numerical servo-
value of 4%. Moreover, the densities of the coarse aggregates mechanism within PFC2D. During the simulation, the axial
and asphalt mastic were 2.7 g/cm3 and 2.0 g/cm3, re- deviatoric stress and axial strain were recorded together for
spectively. By the proposed algorithm, the virtual container the dynamic modulus calculations.
with a height of 150 mm and a width of 100 mm were de- According to the PFC2D manuals [35], several consti-
veloped firstly, then followed by the coarse aggregate, mastic tutive models were used for characterizing the mechanical
and void generation, respectively, within PFC2D as shown in behavior of the heterogeneous materials including the
Figure 12. contact-stiffness models, sliding models, and Burger’s
To verify the precision of the inner components mod- models. The contact force in normal and shear direction
eling, the final generated area of coarse aggregates, asphalt between two entities was determined by the key micro-
mastic, and voids were measured afterwards by the user- parameters in contact-stiffness models, known as the normal
defined routine within PFC2D. The results were summarized stiffness kn and shear stiffness ks . The sliding models were
in Tables 4 and 5. As shown, the error is 5.1%, 4.7%, 2.1%, used for sliding movements by the frictional coefficient while
and 1.7% for the 16 mm, 13.2 mm, 9.5 mm, and 4.75 mm Burger’s models characterized the viscoelasticity of asphalt
aggregates, respectively. The modeling error decreased as the part. To reduce the error of the selected microparameters,
size decreased. This is due to the overflow of the lastly the parameter calibrations were conducted separately for the
generated coarse particles of each grade and it is inevitable. coarse aggregates and asphalt mastic. According to the
When comes to the asphalt mastic, the modeling is signif- Chinese test standards [36], the penetration tests of the
icantly precise with 0.014% error only. The sole difference 4.75 mm, 9.5 mm, 13.2 mm, and 16 mm particles were
between the realistic and virtual mastic content was caused carried out to get the force-displacement curves. Then virtual
by the rounding error in mapping area calculations. simulations for corresponding penetration tests were de-
veloped within PFC2D and were conducted for virtual re-
6. Performance Prediction of the sults. By adjusting the microparameters of different size
Rebuilt Models particles continuously until the simulation curves match the
realistic, the best group of normal stiffness kn , shear stiffness
6.1. Experiments. Based on the Chinese test standards [34], ks , and frictional coefficient could be determined as shown in
uniaxial compressive test was conducted to evaluate the Table 6. Two kinds of contact points were defined by
dynamic modulus of asphalt mixture in laboratory. The Burger’s models. One is the contact point within asphalt,
gradations of mixtures were shown in Table 2. And the and another is between the asphalt mastic and aggregates.
Advances in Civil Engineering 9

Table 7: Microparameters at contacts within asphalt mastic.

Kmn (Pa·m) Kkn (Pa·m) Cmn (Pa·m·s) Ckn (Pa·m·s) Kms (Pa·m) Kks (Pa·m) Cms (Pa·m·s) Cks (Pa·m·s)
1.14e3 7.92e2 1.95e6 5.58e4 3.79e2 2.64e1 6.49e5 1.86e4

Table 8: Microparameters at contacts between asphalt mastic and aggregates.

′ (Pa·m)
Kmn K′kn (Pa·m) ′ (Pa·m·s)
Cmn C′kn (Pa·m·s) K′ms (Pa·m) K′ks (Pa·m) C′ms (Pa·m·s) ′ (Pa·m·s)
Cks
2.27e3 1.58e3 3.89e6 1.12e5 1.51e3 5.28e5 1.30e6 3.72e4

30 7.8%, 5.4% and 4.8% for the 25 Hz, 10 Hz, 5 Hz, 1 Hz, and
26.7
0.1 Hz, respectively. As the loading frequency decreased, the
24.1
25 23.2 22.5 22.7 error decreased obviously. This is because the micro-
Dynamic modulus (GPa)

21.2
parameters for simulations were all calibrated and de-
20 18.517.8
termined by tests under the static loading rather than the
15 13.312.9
cyclic loading. So when the load frequency is low in sim-
ulations, the cyclic loading could be taken as the static
10 loading to some extent. When the load frequency increased,
the difference between the static and cyclic loading grew
5 leading to the error shown in the Figures 13 and 14. In
summaries, the developed models can well predict the dy-
0
25 10 5 1 0.1 namic modulus of asphalt mixture, especially at low fre-
Loading frequency (Hz) quency, which verified the correctness of the proposed
algorithm.
Laboratory test
Simulation
Figure 13: Dynamic modulus prediction based on the proposed 7. Conclusions
models.
This paper proposed an algorithm for the reconstitution of the
virtual asphalt mixture based on the DEM methods. By filling
14 13.1
the shape contours with balls, the virtual coarse aggregates
12.5 were developed precisely. Based on the algorithm for dis-
12 tinguishing the external and internal area of irregular par-
10 9.7 ticles, the asphalt mastic was generated properly by the
9.2
designed rules. In the end, the validation of the proposed
Phase angle (°)

8 6.9 algorithm was verified by the virtual uniaxial compressive test.

6.4 The main conclusions drawn from the study are as follows:
6 5.2
4.7
(1) Virtual shapes were rebuilt precisely based on the
4 3.1 3.5 AIMS scanning. By the random control parameters,
2 numerous virtual particles derived from the standard
ones were developed rapidly with various shapes and
0
25 10 5 1 0.1
sizes. By combining the scanning technology with
Loading frequency (Hz) the stochastic algorithm, this method can well meet
the DEM simulation needs.
Laboratory test
Simulation (2) The virtual asphalt mixtures were rebuilt in three
steps. Firstly, the virtual coarse aggregates were
Figure 14: Phase angle prediction based on the proposed models. distributed randomly to form the skeletons. Then an
innovate algorithm was proposed to distinguish the
Based on the previous studies [37], the microparameters of external and internal area of the coarse particles. At
Burger’s models at different contacts were shown in Tables 7 last, by filling the designed area with uniform
and 8. arranged balls and deleting mastic balls randomly,
Figure 13 shows the simulation results compared to the the asphalt mastic and void could be generated
laboratory tests at the test temperature of 0°C. As shown, successfully. The results shows that the proposed
the virtual results have a good accordance with the truth. The algorithm can well model the inner components
prediction error of dynamic modulus is 12.7%, 6.7%, 6.6%, consistent to the actual gradations, asphalt content,
3.8%, and 3.0% for the 25 Hz, 10 Hz, 5 Hz, 1 Hz, and 0.1 Hz, and voids.
respectively. As shown in Figure 14, the predictions of phase (3) The validation of the developed models was verified
angle is also effective with the minor error of 12.9%, 10.6%, by the virtual uniaxial compressive test. The error of
10 Advances in Civil Engineering

dynamic modulus is 12.7%, 6.7%, 6.6%, 3.8%, 3.0%, aggregate degradation,” Construction and Building Materials,
and the error of phase angle is 12.9%, 10.6%, 7.8%, vol. 22, no. 5, pp. 972–980, 2008.
5.4%, and 4.8% for the 25 Hz, 10 Hz, 15 Hz, 1 Hz, and [12] I. Isailović, M. P. Wistuba, and A. C. Falchetto, “Investigation
0.1 Hz, respectively. It is indicated that the dynamic on mixture recovery properties in fatigue tests,” Road Ma-
modulus of the asphalt mixtures can be well pre- terials and Pavement Design, pp. 1–11, 2017.
[13] A. Hasan, Void Effects on Fatigue Life Asphaltic Concrete,
dicted based on the proposed algorithm, especially at
1973, http://arizona.openrepository.com/arizona/bitstream/
low frequency. 10150/554485/1/AZU_TD_BOX101_E9791_1973_160.pdf.
[14] L. Li, L. Cao, and Y. Guo, “Influence of initial air void on
Conflicts of Interest performance of asphalt mixtures,” Journal of Tongji Univer-
sity, vol. 34, no. 6, p. 757, 2006.
The authors declare that they have no conflicts of interest. [15] J. Hinrichsen and J. Heggen, “Minimum voids in mineral ag-
gregate in hot-mix asphalt based on gradation and volumetric
Acknowledgments properties,” Transportation Research Record: Journal of the
Transportation Research Board, vol. 1545, no. 1, pp. 75–79, 1996.
The study is financially supported by the Science Founda- [16] C. G. Rocco and M. Elices, “Effect of aggregate shape on the
tions of Nanjing Institute of Technology, China mechanical properties of a simple concrete,” Engineering
(YKJ201423), National Natural Science Foundation of China Fracture Mechanics, vol. 76, no. 2, pp. 286–298, 2009.
(no. 51378006), and Natural Science Foundation of Jiangsu [17] L. Liu, D. Shen, H. Chen, and W. Xu, “Aggregate shape effect
Province, China (BK20161421). on the diffusivity of mortar: a 3D numerical investigation by
random packing models of ellipsoidal particles and of convex
polyhedral particles,” Computers and Structures, vol. 144,
References pp. 40–51, 2014.
[1] B. Lira, D. Jelagin, and B. Birgisson, “Gradation-based [18] M. C. Li and I. Kett, “Influence of coarse aggregate shape on
framework for asphalt mixture,” Materials and Structures, the strength of asphalt concrete mixtures,” Highway Research
vol. 46, no. 8, pp. 1401–1414, 2013. Record, vol. 6, pp. 1062–1075, 2005.
[2] I. A. Johan, “The effect of different filler materials and gra- [19] P. A. Cundall, Ball-A Program to Model Granular Media using
dation onrutting and cracking of bituminous mixture,” the Distinct Element Method, Technical Note, Advanced
Journal of Fluid Mechanics, vol. 599, no. 599, pp. 441–464, Technology Group, Dames & Moore, London, UK, 1978.
2008. [20] B. K. Mishra, “A review of computer simulation of tumbling
[3] O. Xu, Z. Wang, and R. Wang, “Effects of aggregate gradations mills by the discrete element method: part II—practical ap-
and binder contents on engineering properties of cement plications,” International Journal of Mineral Processing,
emulsified asphalt mixtures,” Construction and Building vol. 71, no. 1–4, pp. 73–93, 2003.
Materials, vol. 135, pp. 632–640, 2017. [21] K. Huang, T. Xu, and G. Li, “Heating effects of asphalt
[4] D. H. Shen, M. F. Kuo, and J. C. Du, “Properties of gap- pavement during hot in-place recycling using DEM,” Con-
aggregate gradation asphalt mixture and permanent de- struction and Building Materials, vol. 115, pp. 62–69, 2016.
formation,” Construction and Building Materials, vol. 19, [22] A. R. Abbas, Simulation of the Micromechanical Behavior of
no. 2, pp. 147–153, 2005. Asphalt Mixtures using the Discrete Element Method, Washington
[5] J. Zhang, H. Liu, P. Wang, J. Pei, D. Bao, and L. Jin, “Eval- State University, Pullman, WA, USA, 2004.
uation of aggregate gradation and distributing homogeneity [23] H. Yu and S. Shen, “Impact of aggregate packing on dynamic
based on the images of asphalt mixture,” Road Materials and modulus of hot mix asphalt mixtures using three-dimensional
Pavement Design, vol. 18, no. S3, pp. 119–129, 2017. discrete element method,” Construction and Building Mate-
[6] H. Shi, T. Xu, and R. Jiang, “Combustion mechanism of four rials, vol. 26, no. 1, pp. 302–309, 2012.
components separated from asphalt binder,” Fuel, vol. 192, [24] J. Chen, B. Huang, and X. Shu, “Air-void distribution analysis
pp. 18–26, 2017. of asphalt mixture using discrete element method,” Journal of
[7] J. Zhang, G. Liu, C. Zhu, and J. Pei, “Evaluation indices of Materials in Civil Engineering, vol. 25, no. 10, pp. 1375–1385,
asphalt–filler interaction ability and the filler critical volume 2013.
fraction based on the complex modulus,” Road Materials and [25] M. Lu and G. R. Mcdowell, “The importance of modelling
Pavement Design, vol. 18, no. 6, pp. 1338–1352, 2017. ballast particle shape in the discrete element method,”
[8] H. Y. Liu and J. L. Dai, “Effect of different gradations on Granular Matter, vol. 9, no. 1-2, pp. 69–80, 2007.
asphalt mixture resistance to rut,” Journal of Changan Uni- [26] N. Das, S. Thomas, J. Kopmann et al., “Modeling granular particle
versity, vol. 24, no. 5, pp. 11–15, 2004. shape using discrete element method,” Monthly Notices of the
[9] E. Coleri, R. Wu, J. M. Signore, and J. Harvey, “Rutting of Royal Astronomical Society, vol. 420, no. 4, pp. 3368–3380, 2012.
rubberized gap-graded and polymer-modified dense-graded [27] D. Zhang, X. Huang, and Y. Zhao, “Algorithms for generating
asphalt overlays in composite pavements,” Transportation three-dimensional aggregates and asphalt mixture samples by
Research Record: Journal of the Transportation Research the discrete-element method,” Journal of Computing in Civil
Board, vol. 2304, no. 1, pp. 195–204, 2012. Engineering, vol. 27, no. 2, pp. 111–117, 2013.
[10] S. Gokhale, B. Choubane, T. Byron, and M. Tia, “Rut initiation [28] E. Masad, “X-ray computed tomography of aggregates and
mechanisms in asphalt mixtures as generated under asphalt mixes,” Materials Evaluation, vol. 62, no. 7,
accelerated pavement testing,” Transportation Research Re- pp. 775–783, 2004.
cord: Journal of the Transportation Research Board, vol. 1940, [29] Z. Zhou, X. Gu, Q. Li, F. Ni, and R. Yuan, “Use of rejuvenator,
no. 1, pp. 136–145, 2005. styrene-butadiene rubber latex, and warm-mix asphalt
[11] G. D. Airey, A. E. Hunter, and A. C. Collop, “The effect of technology to achieve conventional mixture performance
asphalt mixture gradation and compaction energy on with 50% reclaimed asphalt pavement,” Transportation
Advances in Civil Engineering 11

Research Record: Journal of the Transportation Research

Board, vol. 2575, pp. 160–167, 2016.
[30] Q. Li, H. Yang, X. Ma, and F. Ni, “Evaluation of micro-
structure and damage evolution for asphalt pavements in an
advanced repeated load permanent deformation test using
X-ray computed tomography,” Road Materials and Pavement
Design, vol. 18, no. 5, pp. 1135–1158, 2017.
[31] Y. Fu, L. Wang, M. T. Tumay, and Q. Li, “Quantification and
simulation of particle kinematics and local strains in granular
materials using X-ray tomography imaging and discrete-
element method,” Journal of Engineering Mechanics,
vol. 134, no. 2, pp. 143–154, 2008.
[32] C. Jin, X. Yang, and Z. You, “Automated real aggregate
modelling approach in discrete element method based on
X-ray computed tomography images,” International Journal
of Pavement Engineering, vol. 18, no. 9, pp. 837–850, 2015.
[33] Ł. Skarżyński, M. Nitka, and J. Tejchman, “Modelling of
concrete fracture at aggregate level using FEM and DEM
based on X-ray μCT images of internal structure,” Engineering
Fracture Mechanics, vol. 147, pp. 13–35, 2015.
[34] Ministry of Transport of the People’s Republic of China,
Standard Test Methods of Bitumen and Bituminous Mixtures
for Highway Engineering, Ministry of Transport of the People’s
Republic of China, Beijing, China, 2010.
[35] Itasca Consulting Group, PFC2D Manual Version 4.0, Itasca
Consulting Group, Minneapolis, MN, USA, 2008.
[36] Ministry of Transport of the People’s Republic of China, Test
Methods of Soils for Highway Engineering, Ministry of
Transport of the People’s Republic of China, Beijing, China,
2007.
[37] T. Ma, D. Zhang, Y. Zhang, Y. Zhao, and X. Huang, “Effect of
air voids on the high-temperature creep behavior of asphalt
mixture based on three-dimensional discrete element mod-
eling,” Materials and Design, vol. 89, pp. 304–313, 2016.
 International Journal of

Rotating Advances in
Machinery Multimedia

The Scientific
Engineering
Journal of
Journal of

Hindawi
World Journal
Hindawi Publishing Corporation Hindawi
Sensors
Hindawi Hindawi
www.hindawi.com Volume 2018 http://www.hindawi.com
www.hindawi.com Volume 2018
2013 www.hindawi.com Volume 2018 www.hindawi.com Volume 2018 www.hindawi.com Volume 2018

Journal of

Control Science
and Engineering

Advances in
Civil Engineering
Hindawi Hindawi
www.hindawi.com Volume 2018 www.hindawi.com Volume 2018

Submit your manuscripts at

www.hindawi.com

Journal of
Journal of Electrical and Computer
Robotics
Hindawi
Engineering
Hindawi
www.hindawi.com Volume 2018 www.hindawi.com Volume 2018

VLSI Design
Advances in
OptoElectronics
International Journal of

International Journal of
Modelling &
Simulation
Aerospace
Hindawi Volume 2018
Navigation and
Observation
Hindawi
www.hindawi.com Volume 2018
in Engineering
Hindawi
www.hindawi.com Volume 2018
Engineering
Hindawi
www.hindawi.com Volume 2018
Hindawi
www.hindawi.com www.hindawi.com Volume 2018

International Journal of
International Journal of Antennas and Active and Passive Advances in
Chemical Engineering Propagation Electronic Components Shock and Vibration Acoustics and Vibration
Hindawi Hindawi Hindawi Hindawi Hindawi
www.hindawi.com Volume 2018 www.hindawi.com Volume 2018 www.hindawi.com Volume 2018 www.hindawi.com Volume 2018 www.hindawi.com Volume 2018