See

discussions, stats, and author profiles for this publication at: https://www.researchgate.net/publication/259998427

Creep and recovery behavior characterization of

asphalt mixture in compression
Article in Construction and Building Materials March 2014
DOI: 10.1016/j.conbuildmat.2013.12.088

CITATIONS

READS

151

3 authors, including:
Xinhua Yang

Guowei Zeng

Huazhong University of Science and Technol

Wuhan University of Science and Technology

46 PUBLICATIONS 349 CITATIONS

13 PUBLICATIONS 33 CITATIONS

SEE PROFILE

All in-text references underlined in blue are linked to publications on ResearchGate,

letting you access and read them immediately.

SEE PROFILE

Available from: Xinhua Yang

Retrieved on: 19 November 2016

Construction and Building Materials 54 (2014) 504511

Contents lists available at ScienceDirect

Construction and Building Materials

journal homepage: www.elsevier.com/locate/conbuildmat

Creep and recovery behavior characterization of asphalt mixture

in compression
Fan Bai, Xinhua Yang , Guowei Zeng
School of Civil Engineering and Mechanics, Huazhong University of Science and Technology, Wuhan 430074, Hubei, China

h i g h l i g h t s

g r a p h i c a l a b s t r a c t

 The creep and recovery behaviors of

the asphalt mixture are tested.

 The asphalt mixture exhibits obvious

nonlinearity.
 The simplied Schaperys viscoelastic

model is used for recoverable strain.

 The modied Schwartzs viscoplastic

model is used for irrecoverable strain.

a r t i c l e

i n f o

Article history:
Received 9 September 2013
Received in revised form 21 December 2013
Accepted 25 December 2013

Keywords:
Viscoelastic
Viscoplastic
Asphalt mixture
Creep and recovery behavior
Nonlinear

a b s t r a c t
A simplied Schaperys nonlinear viscoelastic model and a modied Schwartzs viscoplastic model with a
proposed stress function are combined to characterize viscoelastic and viscoplastic behaviors of compressed asphalt mixture. In this combined model, the simplied Schaperys model is used for describing
the recoverable strain, and the modied Schwartzs model is for the irrecoverable strain. The model
parameters are determined with the multiple-stress repeated creeprecovery test and the one-hour
creep test, and then the combined model is validated with some other tests, including the one-hour creep
test at a different stress, the xed-stress repeated creeprecovery test and the random loading test. Compared with the Perzynas viscoplastic model, the modied Schwartzs model can t the experimental
viscoplastic strain response better in a wide stress range, so that the combined model can predict the
creep and recovery behaviors of asphalt mixture under uniaxial compression well.
2014 Elsevier Ltd. All rights reserved.

1. Introduction
Due to permanent deformation accumulation caused by repeated trafc loads, rutting is one of the major failure modes in asphalt concrete pavements. It is extremely important to assess and
improve the resistance of asphalt concrete to permanent deformation before making a desirable design. The laboratory wheel tracking test is the most direct and frequently-used method to
Corresponding author. Tel.: +86 27 87540153.
E-mail address: yangxinh@hust.edu.cn (X. Yang).
0950-0618/$ - see front matter 2014 Elsevier Ltd. All rights reserved.
http://dx.doi.org/10.1016/j.conbuildmat.2013.12.088

investigate the rutting resistance, but it is empirical and cannot

give a deep understanding of the material properties. It has been
well known that asphalt concrete is a time-dependent material
which exhibits noticeable viscoelastic and viscoplastic characteristics [1,2]. These characters can be reected by a creeprecovery
test [3]. When asphalt concrete is subjected to a constant load,
deformation occurs and develops with time. After the load is removed, the deformation can partly recover. To clarify how much
the irrecoverable part is in the total strain, Ossa et al. [4,5] proposed an empirical method to measure the ratio of the recoverable
part to the total axial strain. They found the ratio to be a constant

505

F. Bai et al. / Construction and Building Materials 54 (2014) 504511

which was independent of the total axial strain and scarcely inuenced by the conning pressure. Some other researchers preferred
to decompose the total strain into a viscoelastic strain and a viscoplastic strain and to model them separately [612]. The viscoelastic strain described by a viscoelastic model was used to represent
the recoverable part of total strain, while the viscoplastic strain described by a viscopalstic model was used to describe the irrecoverable part.
The purpose of this paper is to nd a simplied method to
investigate the viscoelastic and viscoplastic characteristics of asphalt mixture in compression. In this method, proper models are
used to describe the mechanical behaviors, and as few experiments as possible are carried out to determine the model
parameters.
There have been many methods to model the viscoelastic strain.
When the strain is very tiny, the Boltzmann superposition principle
is often applicable, and the linear viscoelastic theory can be used
[13]. However, in most cases, especially at high stress and high
temperature, asphalt mixture exhibits obvious nonlinearity with
respect to load because of its complicated composition. Schaperys
nonlinear viscoelastic model [14], originated from the Boltzmann
superposition integral by introducing four stress-dependent
parameters, is still widely used today. The other useful method
to model nonlinear viscoelasticity is the pseudo strain and stress
theory [11,1517]. It can model viscoelasticity as nonlinear elasticity by applying the correspondence principle to eliminate the time
dependence. Theoretically, the strain could be partially recoverable
in a viscoelastic model because a plastic component is often included. For example, a free dashpot can produce plastic strain in
the Maxwell spring-dashpot model [18]. So complete recoverability must be imposed on the viscoelastic model, if it is used to describe the recoverable strain only.
To model the irrecoverable strain, many viscoplastic models
are expressed in differential formulas which dene the viscoplastic strain rate as a function of the current stress and internal variables. Perzynas viscoplastic model is one of them and
has been wildly used [6,19,20]. It assumes that only when
the stress exceeds a threshold can the permanent strain be
produced. The threshold corresponds to the yield surface in
the three-dimensional space, so the model can be extended
into three-dimensional cases easily. However, sometimes the
stress threshold is too low to measure, especially at high temperature cases. Different from Perzynas viscoplastic model,
Schwartzs viscoplastic model [8] uses a stress function to describe complex nonlinear relationship between stress and
viscoplastic strain rate without the necessity of dening a
stress threshold. In addition, recently, a new viscoplastic model
based on the convolution integral was proposed by Subramanian et al. [9]. This model is expressed in an integral formula
and hypothesizes that the viscoplastic strain rate depends not
only on the current values of stress and viscoplatic strain,
but also on the loading history. No matter which formula is
used, a viscoplastic model must embody the strain hardening
effect of the material, and ensure the viscoplastic strain rate
to be zero when the current stress vanishes.
This paper combines a simplied Schaperys nonlinear viscoelastic model and a modied Schwartzs viscoplastic model
with a proposed stress function to characterize viscoelastic
and viscoplastic behaviors of compressed asphalt mixture. The
model parameters are determined with the multiple-stress repeated creeprecovery test and the one-hour creep test. The
multiple-stress repeated creeprecovery test can exhibit the
nonlinear viscoelasticity and viscoplasticity, while the creep
test can reect the long-term response to loading. Both the
strain and stress in this paper are compressive but measured
in positive sign.

2. Combined viscoelasticviscoplastic model

2.1. Strain decomposition
Fig. 1 schematically shows a typical viscoelasticviscoplastic response in a creeprecovery experiment. Before the recovery point
t0 , the asphalt mixture sample creeps because of being subjected
to a constant load. The creep strain can be decomposed into a
recoverable part and an irrecoverable part, which will be described
by the viscoelastic and viscoplastic models, respectively. The
decomposition of the strain can be expressed as

e eve evp ;

where eve and evp denote the viscoelastic and viscoplastic strain
components, respectively. At t0 the load is removed instantaneously
and the creep strain begins to be partly recovered. When the recovery time lasts long enough, only the irrecoverable creep strain evp(t0 )
remains. The recoverable creep strain eve at t0  will be totally recovered after t0 . The instantaneous strain decrement at time t0 is dened as Dee = eve(t0 )  eve(t0 +), which is also called the
instantaneous recovered strain. During the recovery period, the
strain is gradually recovered with time elapsing. The strain decrement from time t0 + to t(t > t0 +) is dened as Dev(t) = eve(t0 +)  eve(t),
which is also called the retarded recovered strain. As shown in
Fig. 1, the recoverable creep strain at t0  can be decomposed into
two parts: Dee and Dev(1).
2.2. Simplied Schaperys nonlinear viscoelastic model
The simplied Schaperys nonlinear viscoelastic model is expressed as

eve g 0 D0 r

Z
0

t


DDt  s

dg 2 r
ds;
ds

where D0 and DD(w) are the instantaneous compliance and the

transient compliance, respectively. g0 and g2 are the stress-dependent parameters.
In the creeprecovery test shown in Fig. 1, the viscoelastic
strain response can be theoretically expressed according to Eq.
(2) [14]. When 0+ 6 t 6 t0 , r = r0 (namely a constant creep stress),
so that

eve t g 0 D0 g 2 DDtr0 ;

0+

but when t P t , r = 0, so that

eve t DDt  DDt  t0 r0 g 2 :

Because the viscoelastic strain is totally recoverable, the strain in

Eq. (4) should be zero when t ? 1. To this end, the following Prony
series is used to represent the transient compliance.

DDt

n
X
Di 1  expki t;

i1

Fig. 1. A typical viscoelasticviscoplastic response in a creeprecovery experiment.

506

F. Bai et al. / Construction and Building Materials 54 (2014) 504511

where Di and ki are the ith compliance and retardation time respectively, and n is the term number of Prony series. Using the Prony
series, the retarded recovered strain Dev(t) as shown in Fig. 1 can
be written as [6]

For comparison, the Perzynas viscoplastic model is also given

here. For the uniaxial compressive case, it can be written as

Dev eve t0  eve t

n
X
r0 g 2 Di 1  expki t0 1  expki tr 

where C is the uidity parameter, f is the yield function, and / is the

overstress function which is nonpositive and usually expressed in
the power law form as

i1

where tr is the recovery time from the recovery point t0 and is dened as tr = t  t0 . The instantaneous recovered strain at time t0 is
expressed as

Dee eve t0  eve t0 r0 g 0 D0 :

The parameters of the Prony series and D0 are all determined by

tting Eqs (6) and (7) with a small-stress creeprecovery test in
which both g0 and g2 are assumed to be unity [14]. g0 and g2 are
used to describe nonlinear viscoelastic responses under high stress
levels, and they can be obtained by conducting creeprecovery
tests at various stresses. g0 can be calculated with the aid of Eq.
(7), and g2 by a stress-shift process. Based on Eq. (6), the strain
quantities Dev at different stresses (e.g. r1 and r2) can be related
by a stress shift index, i.e.

br

Dev r1 ; t r r1 g 2 r1
:

Dev r2 ; t r r2 g 2 r2

If the simplied Schaperys nonlinear viscoelastic model is applicable, the loglog curves of retarded recovered strain Dev  tr at different stresses could form a smooth master curve by vertical shift.
Choosing a small reference stress rref at which the viscoelastic
behavior is linear, i.e. g2(rref) = 1, g2 at a larger stress can be obtained with the aid of Eq. (8) by shifting the corresponding curve
to the reference curve. The amount of vertical shift equals log br.
2.3. Modied Schwartzs viscoplastic model
The Schwartzs viscoplastic model is suitable for low stress
threshold cases, such as high temperature and low aggregate content in asphalt mixture. It describes the strain-hardening effect as
the following power form [8].

e_ vp

gr
;
Aepvp

where A and p are material constants, and g(r) is a stress function

which can be elaborately designed according to experiments. In this
paper, the stress threshold is ignored, and the following nonlinear
stress function is proposed.

r
r
gr
exp c <
ru
rm

c


1> ;

10

where c and c are material constants, ru is used to normalize the

stress function and treated as unity, hi is the McCauley bracket dened as hxi = (x + |x|)/2, and rm is the linear stress limit. When r is
less than rm, Eq. (10) is reduced to g(r) = r/ru, so that the viscoplastic strain rate becomes linear with respect to stress. Because
Eq. (9) is singular for zero viscoplastic strain, the solution is difcult
to be numerically implemented. For simplicity, the Schwartzs model is modied by introducing a small quantity d as follows.

e_ vp

r
ru


 c

exp c < rrm  1 >
p

Aevp d

11

e_ vp C/f ;

12

8
f 0
>
< 0;
 M
/f
;
f
>
: r0 ; f > 0

13

where M is a material constant, and r0y is used to normalize the

yield function and assumed to be unity. A widely used yield function is the extended Drucker-Prager yield surface [6,21,22], which
can be written as

f r; evp r  jevp ;

14

where j(evp) is the hardening function. Only when the stress exceeds j(evp), can the viscoplastic strain be generated. Abu Al-Rub
et al. [6] found good agreement between the Perzynas model and
experiments by expressing the hardening function in the following
exponential form.

jevp j0 j1 1  expj2 evp ;

where j0, j1, and j2 denotes the initial yield stress, the saturated
yield stress, and the strain hardening rate, respectively.
Both the parameters C, M, j0, j1, and j2 in the Perzynas
viscoplastic model and the parameters c, c, A, d, and p in the
modied Schwartzs model can be estimated by the following
optimization process, in which the error between the experimental
measurements and the model predictions is minimized [10].

min EC; M; j0 ; j1 ; j2 or Ec; c; A; d; p

1 XK
2
ei  eivpe ;

i1 vpm
K 1
s:t: C; M; j0 ; j1 ; j2 ; c; c; A; p > 0;

0 < d < 1;

16
17

where eivpe denotes the experimental viscoplastic strain at the ith

point, eivpm is the predicted one by the Perzynas viscoplastic model
or the modied Schwartzs model, and K is the number of data
points. The optimization can be carried out with the help of the
lsqcurvet function in the optimization toolbox of matlab software. The predictions are calculated by the numerical methods that
will be introduced in the next section.
2.4. Numerical implementation
In general, the aforementioned constitutive equations can only
be numerically solved. First of all, the time interval [0+, T] is discretized into N  1 segments by the time points t1, t2, . . . , tN (t1 = 0+,
tj+1 > tj, and tN = T). The loading history is assumed to be known
at every time point and denoted by rj = r(tj) (j = 1, 2, . . . , N). The
numerical methods aim to calculate the viscoelastic and the viscoplastic strains at time point tj, namely ejve and ejvp (j = 1, 2, . . . , N).
A recurrence formula proposed by Simo and Hughes [23] is used
to calculate the viscoelastic strain ejve . Substituting Eq. (5) into Eq.
(2) yields
j
ve

n
X
g 0 rj D0 g 2 rj Di
i1

The smaller d is, the faster the viscoplastic strain grows at the early
stage of loading. An appropriate d can be determined according to
the experiments.

15

rj 

n
X
i
Di h t j ;

18

i1

where h(i)(t) is the hereditary integral for the i th Prony series term
and expressed as

507

F. Bai et al. / Construction and Building Materials 54 (2014) 504511

h t

0

expki t  s

dg 2 rs
ds:
ds

19

It is not necessary to integrate Eq. (19) over the whole integral domain at each time point. By using the midpoint rule [23], the following recurrence formula can be obtained to improve calculation
efciency. For j = 1,
i

h t1 g 2 r1 r1 ;

Cycle

Creep stress (MPa)

Creep duration (s)
Recovery duration (s)

0.15
60
500

0.15
120
600

0.15
400
800

0.15
800
1500

20

And for j = 2, 3, . . . , N,
Rt
i
h tj 0j expki tj  s dg 2drss ds
R tj
i
expki Dtj h tj1 tj1
expki tj  s dg 2drss ds

21

 expki Dtj h tj1 expki Dtj =2g 2 tj rtj  g 2 tj1 rtj1 ;

where Dtj = tj  tj1 is the time increment at the jth time step. The
viscoelastic strain ejve is calculated by Eq. (18) with the updated
h(i)(tj) according to Eq. (21).
Both the viscoplastic constitutive Eqs. (11) and (12) can be
classied into the rst-order nonlinear differential equation and
written in the following general form.

e_ vp /r; evp :

22

Based on the modied Eulers method [24], the following numerical

formula is used to solve Eq. (22). For j = 1,

e1vp evp t1 0;

23

And for j = 2, 3, . . . , N,

ejvp ej1
vp

Table 2
Loading conditions of the xed-stress repeated creeprecovery test.

i
Dt j h
j1
j1
urj1 ; ej1
v
p urj ; ev p Dt j urj1 ; ev p
2

24

After both the viscoelastic and viscoplastic strains are numerically

obtained, the total strain of model can be calculated by

ei eive eivp ; i 1; 2; . . . ; N:

maximum random packing densities for various aggregate gradations can be found
in Deshpande and Cebon [26]. Table 5 gives the conventional physical properties of
the bitumen used here.
The specimens are cylindrical and fabricated by a compaction procedure. The
size errors of 50.0 0.5 mm for diameter, 50.0 0.5 mm for height, 219 1 g for
mass, and 4.0 0.5% for porosity rate are allowed. The specimen diameter is demanded to be at least 10 times greater than the largest aggregate particle size. Several cylindrical molds with two cylindrical plungers are manufactured for casting
the specimens. The molds have inner sizes of 50.0 mm diameter and 100.0 mm
height, while the plungers have sizes of 50.0 mm diameter and 25.0 mm height.
At the rst step, both aggregates and bitumen are heated to about 150 C and then
are mixed and stirred into a hot paste of asphalt mixture. After that, 219.0 g of mixture is taken and poured into an oil-lubricated mold, and then two plungers are
used to compact the mixture by a press machine. When the two plungers are just
completely pushed into the mold from the two ends, the designed sample shape
is formed. Finally, after being cooled down to the room temperature slowly, the
sample is taken out from the mold and stored at a subzero temperature.
All the tests are performed in an electronic universal testing machine at the
room temperature. A sample is uniaxially compressed between two at platens.
Two sheets of paper damped with lubricating oil are put on the top and bottom surfaces of the sample to reduce friction between the sample and the platens. The
room temperature is kept at 20 2 C by an air conditioner and monitored by a
temperature detector. Both the displacement and force are measured as a function
of time with a digital recording system with the measurement accuracies of 1 lm
and 0.1 N. The accuracy of force control in creep tests is 1 N.

25

3. Experiments
In this paper, some tests at different loading and unloading conditions are carried out. Among them, a multiple-stress repeated creeprecovery test and a onehour creep test under 0.1 MPa are for determination of the model parameters, while
another one-hour creep test under 0.15 MPa, a xed-stress repeated creeprecovery test and a random loading test are for validation of the proposed model.
The loading conditions of the multiple- and xed-stress repeated creeprecovery tests are given in Tables 1 and 2, respectively. In the multiple-stress repeated
creeprecovery test, the creep duration in each cycle is xed but the creep stress
increases with increasing cycles, and the recovery duration is long enough so that
the recoverable strain can almost be totally recovered. In the xed-stress repeated
creeprecovery test, the creep stress is xed, but the creep duration increases with
increasing cycle number. The loading conditions of the random loading test are given in Table 3. The creep stress, the creep duration and the recovery duration in
every cycle of the random loading test are all random. In the tests, the loads are applied and removed by the controlled-displacement method with the displacement
rate of 15 mm/min, but in the creep processes the constant loads are maintained by
the controlled-force method.
All the samples used in the tests are made of fully dense idealized asphalt mixture. Table 4 shows the mixture composition. The fully dense idealized asphalt mixture may theoretically have the maximum volume fraction of aggregate for a
certain aggregate shape and gradation. For the aggregate gradation in Table 4, the
maximum volume fraction of sphere aggregate is 75% according to particle packing
characteristics [25]. The residual space between the aggregates is lled with 21% asphalt and 4% air voids. More details about the idealized asphalt mixture and the

4. Parameter determination
Asphalt mixture usually exhibits the linear behavior in small
stress or strain cases. Airey and Rahimzadeh [27] found that the
linear strain limitation was about 1% in the dynamic uniaxial tensioncompression tests of asphalt mixture from low to intermediate temperatures with a purpose-build servohydraulic testing
equipment. Cheung and Cebon [28] pointed out that the 50 pen
bitumen at 20 C was approximately linear viscous when the stress
was below 0.1 MPa. Kim et al. [16] also found that asphalt mixture
behaved in a linear viscoelastic manner in uniaxial controlledstress repetitive tests at 25 C when the stress was less than
0.068 MPa. Accordingly, it can be assumed that the idealized asphalt mixture is linear at 20 C and below 0.05 MPa, so that the
stress-dependent parameters g0 and g2 in Eq. (2) can be approximated as 1 when the applied stress is less than 0.05 MPa, and
the linearity limitation rm in Eq. (11) is also treated as 0.05 MPa.
The parameters in the simplied Schaperys viscoelastic model
are determined from the multiple-stress repeated creeprecovery
test. The strain response in this test is shown in Fig. 2. Because
the recovery duration in each cycle is long enough to recover the
viscoelastic strain, it can be said that the viscoelastic strains in different cycles are independent of each other. In other words, the
multiple-stress repeated creeprecovery test can be regarded as
a series of independent single creeprecovery tests at different
stresses if only the viscoelastic response is considered. The instantaneous recovered strains Dee in different cycles are used to

Table 1
Loading conditions of the multiple-stress repeated creeprecovery test.
Cycle

Creep stress (MPa)

Creep duration (s)
Recovery duration (s)

0.05
125
500

0.10
125
800

0.15
125
800

0.30
125
1100

0.60
125
1200

1.20
125
1500

2.40
125
1700

508

F. Bai et al. / Construction and Building Materials 54 (2014) 504511

Table 3
Loading conditions of the random loading test.
Cycle

Creep stress (MPa)

Creep duration (s)
Recovery duration (s)

0.15
60
50

0.2
80
200

0.1
80
80

0.1
150
0

0.3
200
700

0.5
100
300

0.3
250
310

Table 4
The composition of the idealized asphalt mixture.
Component

Density (g/cm3)
Volume fraction (%)

Aggregate (size, mm)

0.150.3

1.182.36

2.693
37.5

2.676
37.5

Bitumen

Air voids

1.025
21.0

4.0

Table 5
Conventional physical properties of the asphalt.
Asphalt

Penetration at 25 C
(0.1 mm)

Ductility at 15 C
(cm)

Softening point
(C)

AH-70

69

>200

47

determine the parameters D0 and g0. Because the creep stress cannot be unloaded instantaneously, so the instantaneous recovered
strain is approximated by subtracting the strain at the end of
unloading from the strain at the start of unloading. An unloading
process costs about 2.5 s in average and are very short in comparison with the whole corresponding recovery process, so the above
approximation is reasonable. It is obtained that D0 = 10.272 GPa1
and g0 is plotted in Fig. 3 as a function of stress. To determine the
parameter g2, the retarded recovered strain curves Dev(tr) in different cycles are shifted vertically in the double-logarithmic coordinate system to form a master curve at the reference stress of
0.05 MPa, which is plotted in Fig. 4. It shows that apart from the
curves at 0.05 MPa and 0.1 MPa which are slightly discrepant from
the constructed master curve, the retarded recovered strain curves
at different stresses are nearly overlapped. This fact demonstrates
the applicability of the simplied Schaperys nonlinear viscoelastic
model. The stress shift indexes are plotted in Fig. 5, which reveals
the nonlinear viscoelasticity of asphalt mixture. For linear viscoelasticity, the stress shift index curve in the double-logarithmic
coordinate system should be a straight line with slope 1. After

the stress shift indexes are obtained, the parameter g2 can be calculated by means of Eq. (8). The following nonlinear function is
used to express the stress-dependent parameters g0 and g2.

logg 0;2

8
<
:


 b
a log rrm
0

r P rm
r < rm

where a and b are positive parameters listed in Table 6, and rm is

the linearity limitation stress and assumed to be 0.05 MPa according to the above linearity assumption. It can be seen from Eq. (26)
that g0,1 = 1 when r < rm. The variation curves of g0 and g2 with
the stress are plotted in Figs. 3 and 6, respectively. The parameters
of the Prony series in Eq. (5), listed in Table 7, are obtained by tting
the retarded recovered strain master curve in Fig. 4 with Eq. (6). In
Fig. 7, the experimental curves of the total recovered strain at different stresses are compared with the predictions from the simplied
Schaperys nonlinear viscoelastic model.
After the parameters of the nonlinear viscoelastic model are
determined, the viscoelastic strain in the creep period of the multiple-stress repeated creeprecovery test can be calculated numerically, and the viscoplastic strain can also be obtained by
subtracting the calculated viscoelastic strain from the total experimental strain. The obtained viscoplastic strain is plotted in Fig. 8.
It can be seen that the asphalt mixture exhibits obvious viscoplastic behaviors even though under the minimum stress level in the
test. As a simplication, the initial yield stress j0 for the Perzynas
viscopalstic model of Eq. (12) is determined by the optimization
process as well as the other viscopalstic parameters, while as mentioned before, the modied Swchartzs viscopalstic model ignores
the stress threshold.

Fig. 3. Stress-dependent nonlinear parameter g0.

Fig. 2. Experimental and predicted strains for the multiple-stress repeated creep
recovery test.

26

Fig. 4. The retarded recovered strain master curve at 0.05 MPa.

509

F. Bai et al. / Construction and Building Materials 54 (2014) 504511

Fig. 5. The stress shift index.

Fig. 7. Total recovered strain at different stresses.

Table 6
The parameters a and b to determine g0 and g2.
Stress-dependent parameter

g0
g2

0.7450
0.3516

0.7340
1.293

Fig. 8. Viscoplastic strain in the multiple-stress repeated creeprecovery test.

Fig. 6. Stress-dependent nonlinear parameter g2.

Table 7
The Prony series parameters.
i

Di (GPa1)
ki (105 s)

106.9
0.2514

24.79
2.514

10.34
25.14

7.359
251.4

11.06
2514

By the optimization process given by Eqs (16) and (17), the data
from the multiple-stress repeated creeprecovery test is sufcient
to determine the viscoplastic parameters. However, the creep
duration at each cycle in the multiple-stress repeated creeprecovery test is only 125 s. In order to include more information about
the long-term viscoplastic characteristics, the results of the onehour creep test at 0.1 MPa shown in Fig. 9 are also utilized. The
model errors in the multiple-stress repeated creeprecovery test
and the creep test at 0.1 MPa are added as the objective function
of the optimization process to obtain the vicoplastic parameters.
The obtained optimal parameters for the Perzynas viscoplastic
model and the modied Schwartzs model are listed in Tables 8
and 9, respectively. The comparisons between the predictions

Fig. 9. Viscoplastic strain in the 0.1 MPa creep test.

Table 8
The parameters of the Pezynas viscoplastic model.

r0y (MPa)

C (s1)

j0 (MPa)

j1 (MPa)

j2

1.000

1.491  104

1.167

7.825  1014

0.1134

44.79

510

F. Bai et al. / Construction and Building Materials 54 (2014) 504511

Table 9
The parameters of the modied Schwartzs viscoplastic model.

ru (MPa)
1.000

rm (MPa)
0.050

c
5.673  10

c
2

A (s)

1.151

1.088  108

d
5

7.004  10

1.899

and the experimental results are shown in Figs. 8 and 9. It can be

seen that the modied Schwartzs viscoplastic model is in better
agreement with the experimental results than the Pezynas viscoplastic model. It is because the modied Schwartzs viscoplastic
model uses a proper nonlinear stress function to express the relationship between the viscoplastic strain rate and the current stress.
So in the next section, the viscoplastic response of the asphalt mixture is predicted by the modied Schwartzs viscoplastic model
rather than the Pezynas viscoplastic model.
5. Validation and discussion

Fig. 10. Experimental and predicted creep strains at 0.15 MPa.

As shown in Fig. 2, the combination of the Schaperys nonlinear

viscoelastic model and the modied Schwartzs viscoplastic model
can well t the multiple-stress repeated creeprecovery test data.
However, it cannot be naturally concluded that the combined model is also valid for other loading conditions unless they can be validated by certain experiments. For the validation purpose, the
strain responses in the 0.15 MPa creep test, the xed-stress repeated creeprecovery test shown in Table 2, and the random loading test shown in Table 3 are predicted by the combined model.
The comparisons between the predictions and the corresponding
experimental results are given in Figs. 1012. It can be seen that
the predictions exhibit consistent trends with the experimental results, although obvious discrepancies can also be found in Figs. 10
and 11. These errors may be partly caused by the discrepancies between samples. In Fig. 12, the prediction agrees well with the
experimental strain, which indicates that the sample used in the
random loading test maybe has a little difference with those used
in the two tests determining the model parameters.
6. Conclusions

Fig. 11. Experimental and predicted strains in the xed-stress repeated creep
recovery test.

Fig. 12. Experimental and predicted strains in the random loading test.

In this study, the uniaxial compressive creep and recovery

behaviors of the idealized asphalt mixture are studied experimentally, and a combined constitutive model is proposed to characterize them. The recoverable strain is described by the simplied
Schaperys nonlinear viscoelastic model, while the irrecoverable
strain is done by the modied Schwartzs viscoplastic model. Based
upon the aforementioned work, the following conclusions can be
made.
(1) Asphalt mixture exhibits obvious nonlinearity. The retarded
recovered strain curves at different stress levels can be
shifted vertically in the double-logarithmic coordinate system to form a smooth master curve, so the simplied Schaperys nonlinear viscoelastic model is accurate enough to
describe the recoverable strain.
(2) The Perzynas viscoplastic model needs a stress threshold to
determine whether the viscoplastic strain can be activated,
while it is not necessary for the Schwartzs visocplastic
model. So the Schwartzs visocplastic model is more suitable
to describe the mechanical behavior of asphalt mixture at
high temperature.
(3) The modied Schwartzs viscoplastic model expresses the
viscoplastic strain rate as a complicated nonlinear function
of the stress, so it can better t the experimental viscoplastic
strain response in a wide stress range than the Perzynas
viscoplastic model.

F. Bai et al. / Construction and Building Materials 54 (2014) 504511

(4) By combining the simplied Schaperys nonlinear viscoelastic model and the modied Schwartzs viscoplastic model,
the creep and recovery behaviors of asphalt mixture under
uniaxial compression can be well predicted.

Acknowledgments
This work is supported by the National Natural Science Foundation of China (Nos. 10872073) and National Basic Research Program of China (973 Program: 2011CB013800).
References
[1] Krishnan JM, Rajagopal KR. Review of the uses and modeling of bitumen from
ancient to modern times. Appl Mech Rev 2003;56(2):149214.
[2] Bai F, Yang XH, Yin AY, Zeng GW. Prediction method of long-term mechanical
behavior of largely deformed sand asphalt with constant loading creep tests. J
Eng Mech-ASCE 2012;138(12):145767.
[3] Drescher A, Kim JR, Newcomb DE. Permanent deformation in asphalt concrete.
J Mater Civil Eng 1993;5(1):11228.
[4] Ossa EA, Deshpande VS, Cebon D. Phenomenological model for monotonic and
cyclic behavior of pure bitumen. J Mater Civil Eng 2005;17(2):18897.
[5] Ossa EA, Deshpande VS, Cebon D. Triaxial deformation behavior of bituminous
mixes. J Mater Civil Eng 2010;22(2):12435.
[6] Abu Al-Rub RK, Masad EA, Huang CW. Improving the sustainability of asphalt
pavements through developing a predictive model with fundamental material
properties. Texas Transportation Institute; 2009 Aug. Contract No. TRT07-G0006. Report No. SWUTC/08/476660-0007-1.
[7] Levenberg E. Viscoplastic response and modeling of asphalt-aggregate mixes.
Mater Struct 2009;42(8):113951.
[8] Schwartz CW, Gibson NH, Schapery RA, Witczak MW. Viscoplasticity modeling
of asphalt concrete behavior. In: 15th Engineering Mechanics Division
Conference; 2002 Jun 4; Columbia University, New York. ASCE; 2004. p.
144159.
[9] Subramanian V, Guddati MN, Kim YR. A viscoplastic model for rate-dependent
hardening for asphalt concrete in compression. Mech Mater 2013;59:14259.
[10] Sun L, Zhu Y. A serial two-stage viscoelastic-viscoplastic constitutive model
with thermodynamical consistency for characterizing time-dependent
deformation behavior of asphalt concrete mixtures. Constr Build Mater
2013;40:58495.

511

[11] Underwood SB, Kim RY. Viscoelastoplastic continuum damage model for
asphalt concrete in tension. J Eng Mech-ASCE 2011;137(11):7329.
[12] Ye Y, Yang X, Chen C. Modied Schaperys model for asphalt sand. J Eng MechASCE 2010;136(4):44854.
[13] Elsei M, Mohammad L, Dessouky S, Yang W, Ying H. A viscoelastic model to
describe the constitutive behavior of unmodied asphalt binders at low
temperatures. In: Pavements and Materials: Characterization and Modeling
Symposium at EMI Conference; 2010 Aug 811; Los Angeles. ASCE; 2010. p. 1
15.
[14] Schapery RA. On the characterization of nonlinear viscoelastic materials.
Polym Eng Sci 1969;9(4):295310.
[15] Kim YR, Little DN. One-dimensional constitutive modeling of asphalt concrete.
J Eng Mech-ASCE 1990;116(4):75172.
[16] Kim YR, Lee YC, Lee HJ. Correspondence principle for characterization of
asphalt concrete. J Mater Civil Eng 1995;7(1):5968.
[17] Lee HJ, Kim YR. Viscoelastic constitutive model for asphalt concrete under
cyclic loading. J Eng Mech-ASCE 1998;124(1):3240.
[18] Knauss WG, Emri I, Lu H. Mechanics of polymers: viscoelasticity. In: Sharpe Jr
WN, editor. Springer handbook of experimental solid mechanics. New
York: Springer; 2008. p. 4996.
[19] Darabi MK, Abu Al-Rub RK, Masad EA, Huang CW, Little DN. A modied
viscoplastic model to predict the permanent deformation of asphaltic
materials under cyclic-compression loading at high temperatures. Int J
Plasticity 2012;35:10034.
[20] Abu Al-Rub RK, Darabi MK, Kim SM, Little DN, Glover CJ. Mechanistic-based
constitutive modeling of oxidative aging in aging-susceptible materials and its
effect on the damage potential of asphalt concrete. Constr Build Mater
2013;41:43954.
[21] Masad E, Tashman L, Little D, Zbib H. Viscoplastic modeling of asphalt mixes
with the effects of anisotropy, damage and aggregate characteristics. Mech
Mater 2005;37(12):124256.
[22] Saadeh S, Masad E, Little D. Characterization of asphalt mix response under
repeated loading using anisotropic nonlinear viscoelastic-viscoplastic model. J
Mater Civil Eng 2007;19(10):91224.
[23] Simo JC, Hughes TJR. Computational inelasticity. New York: Springer; 1998.
[24] Quarteroni A, Sacco R, Saleri F. Numerical mathematics. 2nd ed. New
York: Springer; 2007.
[25] German RM. Particle packing characteristics. Princeton, N.J.: Metal Powder
Industries Federation; 1989.
[26] Deshpande VS, Cebon D. Uniaxial experiments on idealized asphalt mixes. J
Mater Civil Eng 2000;12(3):26271.
[27] Airey GD, Rahimzadeh B. Combined bituminous binder and mixture linear
rheological properties. Constr Build Mater 2004;18(7):53548.
[28] Cheung CY, Cebon D. Experimental study of pure bitumens in tension,
compression, and shear. J Rheol 1997;41(1):4574.