VISCOELASTIC CONSTIlUTlVE MODEL FOR ASPHALT CONCRETE

UNDER CYCLIC LOADING

By Hyun-Jong Lee l and Y. Richard Kim/ Members, ASCE

AeSTRACT: This paper presents a mechanistic approach to uniaxial viscoelastic constitutive modeling of as-
phalt concrete that accounts for damage evolution under cyc lic loading conditions. Schapery's elastic-viscoelastic
correspondence principie is applied as a means of separately evaluating viscoelasticity and time-dependent
damage growth in asphalt concrete. The time-dependent damage growth in asphalt concrete is modeled by using
a damage parameter based on a generalization of microcrack growth law. Internal state variable formulation was
used in developing the analytical representation of the model. Tensile uniaxial fatigue tests were performed
under the controlled-strain mode to determine model parameters. Then, the resulting constitutive equation was
used to predict the stress-strain behavior of the same materials under controlled-stress mode. The constitutive
equation proposed in this paper satisfactorily predicts the constitutive behavior of asphalt concrete all the way
up to failure under various loading conditions including different stress-strain amplitudes, monotonie versus
cyclic loadings, and different modes of loading.

INTRODUCTION CONSTITUTIVE THEORY

Fatigue cracking due to repeated traffic loading is one of Correspondence Principle

the major distresses in asphalt concrete pavements. Accurate Schapery (1984) proposed the extended elastic-viscoelastic
prediction of the fatigue performance of these pavements re- correspondence principle, which can be applicable to both lin-
quires understanding of the hysteretic behavior of asphalt con- ear and nonlinear viscoelastic materials. He suggested that
crete under realistic traffic conditions. constitutive equations for certain viscoelastic media are iden-
The hysteretic behavior of asphalt concrete under traffic tical to those for the elastic cases, but stresses and strains are
loading is attributed to three major mechanisms (Kim and Lit- not necessarily physical quantities in the viscoelastic body. In-
tle 1990; Kim et al. 1995): damage growth, relaxation of stead, they are pseudo variabies. For the case of a growing
stresses in the system because of the viscoelasticity of asphalt traction boundary surface, such as crack growth, the viscoe-
concrete, and microcrack healing. Because these three mech- lastic problem can be reduced to an elastic case by using phy s-
anisms take place simultaneously, it is important to model the ical stresses and pseudo strains
damage growth separately from the viscoelastic effect to sim- Uniaxial pseudo strain is defined as follows:
plify the modeling of the mechanical behavior of asphalt con-
crete. eR =-ERl L'o E(t - T) - ae dT
aT
(1)
Kim and Little (1990) applied the correspondence principie
(Schapery 1984) to transform the viscoelastic analysis to an
where ER = reference modulus that is an arbitrary constant;
elastic case. A time-dependent damage parameter based on the and E(t) = relaxation modulus. For linear viscoelastic materi-
microcrack growth law (Schapery 1981) was then employed aIs, a uniaxial stress-strain relationship is
to model the damage growth in asphalt concrete. The resulting
constitutive model satisfactorily predicted the hysteretic be- ' de
havior of sand-asphalt under the controlled-strain mode. In this
study, the same methodology was used to model the consti-
CI =
L
o
E(t - T) - dT
dT
(2)

tutive behavior of asphaIt concrete. With the use of the definition of pseudo strain in (1), (2) can
Although Kim and Little (1990) successfully modeled the be rewritten as
short-term fatigue damage evolution of sand-asphalt, the ma-
(3)
terials and testing conditions were less representative of the
field conditions. In this study, therefore, more realistic condi- A correspondence can be found between (3) and a linear el as-
tions were used in the modeling to increase the applicability tic stress-strain relationship.
of the model to actual pavement design and analysis. The cy-
clic testing conditions used in Kim and Little (1990) and the Damage Parameter
ones used in this study are presented in Table 1. As can be
seen from this table, the major difference in this study from For viscoelastic materiais, Schapery (1984) showed that the
Kim and Little (1990) is the faet that the mode-of-loading local crack growth rate (da/dt) obeys a power law in local J
effects and long-term fatigue damage evolution of asphalt con- integral
crete are incIuded in the constitutive modeIing.
-da
dt
=A (J v)
k
(4)
Ipost-PhD Res. Assoc., Dept. of Civ. Engrg., North Carolina State
Univ., Raleigh, NC 27695-7908.
2 Assoc. Prof., Dept. of Civ. Engrg., North Carolina State Univ., Ra- where J v = generalized J integral; and A and k =positive con-
leigh, NC. stants. According to the theory, the exponent k is a function
Note. Associate Editor: Robert Y. Liang. Discussion open until June that depends on characteristics of a failure zone. For example,
l, 1998. To extend the cIosing date one month, a written request must if the material's fracture energy (r) and failure stress within
be filed with the ASCE Manager of Joumals. The manuscript for this the fraeture process zone are constants, then k = l + 11m,
paper was submitted for review and possible publication on February 26,
1996. This paper is part of the Journal 0/ Engineerlng Mechanics, Vol. where m is the exponent of creep compliance versus time in
124, No. l, January, 1998. ©ASCE, ISSN 0733-9399/98/0001-0032- the power law relationship. If the fracture process zone size
0040/$4.00 + $.50 per page. Paper No. 12798. and r are constants, k = 11m.
32/ JOURNAL OF ENGINEERING MECHANICS / JANUARY 1998
 TABLE 1. Comparleon ot Testing Condltlons tV 1E-02 r--=_____- - - - - - - - - - - - - - ,
~ • Refel'8nce Temperatul'8: 25C
Parameler Presanl Sludy Kim and Utlle (1990)
(2) (3) :::.. 1 E-03 -5 C - :Beat Fil In MPL
(1)
Mode of loading
Wave fonn
Loading rate
Controlled-strain and -stress
Haversine
0.1 s/cycle
Controlled-strain
Sawtooth
I s/cycle
.g 1E-04
Q.
Sand-asphalt
Mixture type
Time of interest
Densely graded asphalt concrete
Long-tenn (up to failure) Shon-tenn
O
g 1E-05
ł
... 1E-06
The final form of the damage parameter based on the gen- O
eralization of the rnicrocrack growth law is ~ 1E-07

Sp = (L lal/NIP dtfP (5)

li
::::E 1 E-08
1E-6
L--_--''--_---L_ _--I.._ _--'-_ _...J
1E-4 1E-2 1EO 1E2
(a)

1E4
where p = (1 + N)k, and a - leRIN. Reduced Tlme (sec)
To prediet strains from known stresses, the damage param-
eter based on stress in (5) is useful. However, if the strains ~ 1E+08r----------------,
are known, the following damage parameter based on pseudo as Refel'8nce Temperatul'8: 25C
a..
Slrain is recommended:
o;- 1E+07
Sp = (L leRIP dt) l/p (6) ~
:l

::::E
1E+06
c:
This form is known as a Lebesgue norm of eR. When p goes o 1E+05
to infinity (Le., for elastic materials), Sp becomes the largest
absolute value of pseudo strain up to the current time, denoted I 1E+04
•
5C
here by Ief I (Reddy and Rasmussen 1982). ~... o
CI)
25C
MATERIALS AND TESTING PROCEDURE li
1E+03
• •
40C (b)
::::E
AU specimens were fabricated using Watsonville granite ag- 1E+02
gregate and AAD and AAM asphalt cements used in the Stra- 1E-6 1E-4 1E-2 1EO 1E2 1E4
tegie Highway Research Program (hereinafter, denoted as Reduced Tlme (sec)
AAD and AAM mixtures, respectively). Aggregates were
mixed with 5.0% asphalt cement by weight of dry aggregate FIG. 1. Creep and Relaxatlon Master Curves at Reference
Temperature: (a) Creep Master Curve; (b) Relaxatlon Master
at 140° for 4 min. The mixture was compacted in a rectangular Curve
mold at 116°C by a rolling wheel compactor and cured at room
temperature for 2 days before coring 102 mm (4 in.) diameter
Based on the linear viscoelastic stress-strain relationship in
by 203 mm (8 in.) height cylindrical specimens from the slab.
(2), relaxation moduli were predicted from the creep compli-
The specimens were cored in a horizonta1 direction perpen-
ance values as follows:
dicular to the compaction force applied, consistent with the
direction of tensile stress and/or strain developed near the bot- 11,375,164
tom of an asphalt concrete layer in actual pavement systems. AAD: E(~) = 1,586 + ( + _~_)0.6' (7)
AU tests were conducted in a uniaxial tension mode using
a servohydraulic closed-Ioop testing machine. Agluing jig and l 0.0003
a uniaxial testing apparatus were developed to pro vide precise
alignment of a cylindrical specimen with respect to the loading
12,099,300
axis and to reduce eccentric stress concenlration during tests.
A haversine wave with 0.1 loading time was used in cyclic
AAM: E(~) = 700 + ( + _~ _)0.62 (8)

loading. Stresses and strains used in the analysis were nominal 1 0.0018
(average) values.
where E(~) = master relaxation modulus in kilopascals; and ~
DETERMINATION OF VISCOELASTIC = reduced time. The predicted relaxation modulus for the AAD
MATERIAL PROPERTIES mixture was plotted with the measured values in Fig. l(b).
Generally, a good match is observed between the predicted
Creep and relaxation tests were performed on several spec- and the measured data, although the fit becomes poorer at
imens at -5,5, 15, 25, and 33°C to obtain master creep com- times longer than 10 s because of the lirnitation in data ac-
pliance and master relaxation modulus. The individual creep quisition resolution and the effect of logarithmie scale. For the
compliance-time curves at different temperatures were used remainder of the analysis in this paper, the relaxation modulus
to construct the master creep curve at the reference tempera- predicted from the creep test is used.
ture of 25°C as shown in Fig. l(a) (AAD mixture). Modified
power law (MPL) was used to analytically represent the master CALCULATlON OF PSEUDO STRAIN USING
creep curve. SUPERPOSITION
Calculation of pseudo strain requires the expression of re-
laxation modulus as a function of time. The relaxation test is Pseudo strain is an essential parameter for applying Schap-
fairly cumbersorne to perform because the immediate increase ery's (1984) correspondence principIe to the hysteretie stress-
in strain input results in a large initialload response. For prac- slrain behavior of asphalt concrete. As shown in (1), the com-
tieal purposes, it is desirabIe to predict the relaxation modulus putation of the pseudo Slrain requires the acquisition of slrain
from a simpler test, such as the creep test (Kim et al. 1995). values from time zero to time t. This method is impractical to
JOURNAL OF ENGINEERING MECHANICS / JANUARY 1998/33
 be used in the long-term fatigue analysis because of limitations strains demonstrates the validity of this approach for the strain
in computer memory capacity and computing time. An alter- histories analyzed in this study.
nate way to calculate the pseudo strain is to present the relax-
ation modulus and the strain as analytical functions of time CONSTITUTIVE MODELING OF ASPHALT CONCRETE
and integrate the product of these functions. In the controlled-
strain mode, the haversine strain his tory can be represented Experlmantal Study o, Corraspondanca Principia
simply in an analytical function composed of a step loading
and a sinusoidal function. In the controlled-stress mode, the Prior to constitutive modeling, the material's behavior was
average strain and strain amplitude of each cyc le increase with studied with the application of the correspondence principie.
time. Thus, the strain his tory under the controlled-stress mode The types of uniaxial tensile testing used were as follows:
in Fig. 2(a) can be separated into the following two strain
components: l. Controlled-stress cyclic loading tests
I 2. Constant-strain-rate monotonie loading tests with vary-
EI(t) = Bo + BI ta + 2: Blt l; ez(t) = Eo(t)sin(wt + 9) (9, 10) ing strain rates
1.2

where t =time elapsed from the first application of the loading; In the eyelie loading tests, a loading amplitude of 156 N
Eo(t) = amplitude of sinusoidal function, which is a function (35 lbs) was used, which was low enough not to induee any
of time; w = angular velocity; and Bo, Bh BI , l, a, and 9 = signifieant damage. Typical hysteretic stress-strain behavior is
constants. With the use of the superposition property, the presented in Fig. 3(a) at seleeted eyeles. As expected, the
pseudo strain for the strain history presented in Fig. 2(a) can stress-strain loops shift to the right side with the reduction of
be determined by adding the pseudo strain calculated from the dissipated energy determined from the area inside the stress-
first strain component (El) to the pseudo strain calculated from strain eurve. In Fig. 3(b), the same stresses are plotted against
the second strain component (ez). pseudo strains ealeulated based on (1). As ean be seen from
To validate this approach, the measured and the generated Figs. 3(a) and (b), hysteretic behavior due to both loading-
strains are compared in Fig. 2(a). Then, pseudo strain values unloading and repetitive loading has disappeared using the
were calculated for the entire loading his tory from the mea- pseudo strains. AIso, it is noted that the stres s-pseudo strain
sured strains and from the generated strains using the super- behavior in Fig. 3(b) is linear.
position principie. The reference modulus (ER) value of l was In the monotonie loading tests, four different strain rates
used in this study. A typical stress-pseudo strain curve is pre- ranging from 0.00005 to 0.0032 units/s were used. Fig. 4(a)
sented in Fig. 2(b). A good agreement between the pseudo displays the rate-dependent behavior observed from the AAD
strains from the measured strains and from the generated mixture. Pseudo strains were plotted against er/I in Fig. 4(b).

0.0003 25~--------------------------~~~

~
(a)
SlraJn I 1st

0.0002
Gene
20
:'~
. :'j
l 2nd

,'. r; I
5th

~ 0.0001
((. I 10th

!/ I 20th
o 5 ( /
I
) (a)
O~~~~~~~~~~~~~--~~
-0.0001
o 0.1 0.2 0.3 0.4 0.5 o 1E-05 2E-05 3E-05 4E-05 5E-05
Tlma (sac) Strain

1~[8
25.-----------------------------~

Meas
20
Gene
'iil 100 m
a.. ~ 15
~

i
Ci) 50
g
12
en
10

5
(b) (b)
o
o 50 100 150 200 5 10 15 20
Pseudo Strain Pseudo Straln
FIG. 2. Verlflcatlon of Calculatlon Method for Pseudo Stralns FIG. 3. Appllcatlon of Correspondence Prlnclple to CyclIc
In Long-Term Fatlgue Tests: (s) Straln-Tlme; (b) Stress-Pseudo Data wlth Negllglble Damage: (a) Stress-Straln; (b) Stress-
Straln Pseudo Straln

34/ JOURNAL OF ENGINEERING MECHANICS / JANUARY 1998

 800 800

[8
Increase in the
Straln Rates No. ol cycles
(unltlsec)
600 Pred CC 400
0.0032
as
n..
~
400 0.0018
l?
a..
e. +
I CI)
CI)
O
CiS
200
0.0004
-Q)
Lo

CI)

-400
0.0001
(a)
(a)
0.003 0.006 0.009 0.012 -800
Straln O 0.0005 0.001 0.0015 0.002
Strain
1200
S1rain Rates
•
.0001/8
800

ł
900 •
as
n..
.0004Is cC
CC ci"
o CC 400
~ CC
.-...
.001818 CC
600
~
«I
CI CC a..
e.
g ~
.0032Is
CI) CI) O
CI)
300

(b)
-Q)
Lo

CI)

-400
o
o 400 800 1200 1600 (b)
Pseudo Strain -800
FIG. 4. Applleatlon of Correspondenee Prlnelple to Monotonie -1000 -500 O 500 1000
Data: (a) Stress-Straln; (b) Stre88-Pseudo Straln Pseudo Strain
FIG. 5. Cyelle Behavlor ot Asphalt Conerete under Controlled-
The initial pseudo stiffness l, defined as the ratio of stress to Straln Mode: (a) (J' - E; (b) (J' - ER
pseudo strain at the early linear part of a stress-pseudo strain
curve, was necessary to reduce the effect of sample-to-sample
variability. The curves representing different loading rates falI linear because of the damage. In addition, the stiffness reduc-
on the same line at lower stress levels, and then the discrep- tion of the material due to damage growth results in the change
ancy in the stress-pseudo strain curves among the different of the slope of each cycle as cyclic loading continues. To rep-
loading rates becomes greater. This behavior implies that the resent this change of the slope, secant pseudo stiffness, de-
correspondence principIe can successfully eliminate the rate noted by SR, was defined as the ratio of a stress value to a
dependency of the material when the damage is negligible. pseudo strain value at the peak pseudo strain of each cycle.
However, if a significant level of damage is induced in the As can be seen from the loading and unloading curves in
sample, additional variabIes should be employed in a consti- Fig. 5(b), there are two different stresses associated with one
tutive equation to represent the damage growth in the system. f!1 value. On the loading path, the stress is a function of current
f!1 only, whereas on the unloading path, the stress is a function
Uniaxial Viscoelastic Constitutive Model under of current f!1 and the largest f!1 during the f!1 history up to that
Controlled-Strain Mode time (ef). This observation suggests the use of eR/ef in iden-
tifying whether the point of interest is on the loading path or
Because a linear relationship between the stress and pseudo on the unloading path (Kim and Little 1990). That is, eR/ef =
strain has been found when the damage was negligible in the 1 during the loading path, and eR/ef < 1 during the unloading
proceeding section, it was assumed that the material was lin- path. Therefore, (11) becomes
early viscoelastic and that any deviation from the linear vis-
coelastic behavior was caused by damage. Thus, the following
fonn of the constitutive model was employed in this study: CI'
R (eRef' Sm)
=I(e)D (12)

(11)
where m = 2, 3, ... , M.
where l = initial pseudo stiffness; D(Sm) = damage function; It was found from the experimental data that the si ze of
and Sm = internal state variabIes. each stress-pseudo strain hysteresis loop was dependent upon
To study the damage growth in asphalt concrete, the con- the stress amplitude incurred in a specimen. That is, the larger
trolled-strain fatigue tests were conducted with two ampli- the stress amplitude, the bigger the hysteresis loop. Because
tudes. These strain levels were high enough to induce some the stress is the parameter to be predicted from the constitutive
damage to the specimens. Fig. 5 show s typical stress-strain equation and CI' - IeR IN, the amplitude of pseudo strain, de-
and stress-pseudo strain hysteresis loops at different numbers noted by ~, is selected as another internal state variable (lSV)
of cycles. Different from the negligible damage case in Fig. to represent the effect of the stress amplitude on the hysteresis
3(b), the loading and unloading paths in each cycle are non- loops
JOURNAL OF ENGINEERING MECHANICS / JANUARY 1998/35
 Little 1990), P = 2(1 + 11m) was assumed. Typical m values
CJ' =[(e)D (eR
R
ef' EO, S,. R ) (13) obtained from the creep tests were 0.628 and 0.578 for the
AAD and AAM mixtures resulting in the p values of 5.19 and
where m = 3, 4, ... , M. 5.46, respectively.
Although eR/ef and e~ are sufficient to model the hysteretic In Fig. 6(a), ~/l was plotted against Sp determined from (6).
behavior in each cycle, another ISV is needed to model the The data presented in this figure include the data from the
change in the slope of each cycle as cyclic loading continues. entire loading path of the fust cycle and the values of the
Knowing the reduction in this slope is related to the fatigue remaining cycles up to failure. The discrepancies between two
crack growth, the damage parameter Sp is employed in the strain-Ievel data in the first loading path could be caused by
constitutive model as another ISV. Thus (13) becomes experimental errors where some minor adjustrnents occur
within the test setup. However, after the fust loading path, all
CJ' =[(eR)D( Sp, :;, e~) (14) the data from two different strain amplitudes falI on the same
curve up to failure. Because the same loading time (0.1 s) for
To obtain the explicit form of the damage function D, sev- the two strain amplitudes produced different strain rates for
eral forms of the constitutive model were attempted using the the two tests, Sp demonstrates its ability of accounting for the
experimental data, and the following additive form was found effect of strain rates on damage growth. The regression anal-
to yield the best result: ysis on the controlled-strain data resulted in the following
equation for the function F:
CJ' =[(eR) [F(Sp) + G (e~, :;) ] (15) (16)

The regression coefficients in (16) for the two mixtures are in

The function F represents the change in the secant pseudo
Table 2.
stiffness (~) during cyclic loading, and the function G ac-
To find the explicit form of G, F(Sp) was subtracted from
counts for the hysteretic behavior of stres s-pseudo strain re-
lationship.
CJ'/(Ie R ) (hereinafter, called back-calculated G) and then plotted
against eR/ef. As shown in Fig. 6(b), a stron g power relation-
ship was observed between the back-calculated G and eR/ef.
Characterizatlon of Material Functions F and G In this figure, the value of eR/ef increases during loading (load-
To calculate Sp, the parameter p should be determined a ing I for the negative values of eR/ef and loading II for the
priori. In this study, it is assumed that the material is linearly positive values of eR/ef) while it decreases during unloading
viscoelastic (Le., N = l). If so, as discussed earlier, the value (unloading I for the positive values of eR/ef and unloading II
of p is equal to 2(1 + 11m) or 21m depending on the charac- for the negative values of eR/ef). The study of this data pro-
teristics of the failure zone. Based on earlier work (Kim and duced the following form of function G:

2.0 G = Go + Gl 1:;IG,<.g) (17)

c
:::::::
fi) .OOO7unit
The regression coefficients Go and Gl are summarized in Table
fi)
Q)
c 1.5
• 3 for the two mixtures. Because the size of the stress-pseudo
.OOt4unit
= strain hysteresis loop is dependent on the amplitudes of pseudo
~
o F(Sp) strain incurred in the specimens, the exponent G2 in (17) is
'O
::J 1.0 function of e~ instead of constant values.
Q)
Ali the values of G2 at different strain amplitudes were ob-
-
fi)
a.. tained from the regression analysis on the experimental data.
c
lU
(,)
0.5 Because there were four values of G 2 depending on the loading
Q)
CI) • zone for one strain-amplitude data, they were normalized
o (a) through division by a reference value of G2 , denoted by G 2R ,
0.0 for each loading zone. The values of G2 at the strain amplitude
O 500 1000 1500 2000 2500 of 0.0014 units for both mixtures in Table 3 were selected as
Sp
TABLE 2. Coafflclents In Eq. (16)
5 Mixture typa Fo F, F. Boundary condltlon
c (1 ) (2) (3) (4) (5)
~
.OOO7unit AAD 1.008 0.00046 0.93 Sp :S 1,820
3 •
i . .OO14unit AAM
3.174
1.0796
0.00146
0.02612
1.0
0.39
Sp> 1,820
Sp :S 2,000

.,."
Unloading II ; Loading II
i(;l 2.892 0.00119 1.0 Sp> 2,000
'S
o
iii
o
-::..,.
-"1r'IfłD ...
.. :,..,- • u.
G.I/: ....~ TABLE 3. Coefflclants In Eq. (17)
~ -1
o(\I
.c
•
LOading~ oc
. It ~nlOading I Mixtura typa Zona Go G, G2R ~
C3' -3 ~~ (1 ) (2) (3) (4) (5) (6)
.. AAD Loading I
Loading II
0.1917
-0.0933
-0.1765
0.0912
-1.00
-1.30
0.889
0.949
(b) Unloading I 0.1573 -0.1653 -1.00 0.947
-5 Unloading II -0.6945 -0.63 0.846
0.5713
-1 -0.5 o 0.5 AAM Loading I 0.5684 -0.3789 -0.75 0.917
Pseudo Strain/Max. Pseudo Strain Loading II -0.3025 0.2526 -0.80 0.973
Unloading I 0.0008 -0.0760 -1.20 0.976
FIG. 6. Oamaga Evolutlon In Asphalt Concrata undar Con- Unloading II -0.3848 0.2481 -0.90 0.828
trollad-Straln Moda: (a) S"'-Sp; (b) Back-Calculated G-ER/E~

36/ JOURNAL OF ENGINEERING MECHANICS / JANUARY 1998

 1.2 300,------------------------------,
Increase in the
No. ol cycles
0.9 ~
~ (i1 200
i~ 0.6
)(

LVE Umil
a..
e.
CI)
E CI CI)
~
o .OOO7unit ~
Z
• Ci) 100
0.3
.OO14unit

BasI Fil
0.0 (a)
O 500 1000 1500 2000 OL-------~------~------~------~

Pseudo Strain Amplltude o 0.002 0.004 0.006 0.008

Strain
FIG. 7. Normallzed G,. ver.u. E: for AAM Mlxture
TABlE 4. Coefflclentaln Eq. (18) 300,------------------------------.

Mixtura typa ~o ~, Boundary condition

(1) (2) (3) (4)
AAD 0.0 0.0 ef s 200
1.246 -247.066 ef> 200 (i1 200
0.0 0.0 ef s 200 a..
AAM
1.155 -230.606 ef> 200 e.
CI)
CI)

~
the G 2R values. The nonnalized G2 values were then plotted Ci) 100
against e~ for the AAM mixture in Fig. 7. As shown in this
figure, it was desirable to have some more data near the origin
to obtain regression curves because only two strain-ampIitude
data were available for each mixture. As discussed earIier, no o~----~----~----~----~----~
stress-pseudo strain hysteresis loop developed (i.e., G2 = O) o 100 200 300 400 500
when the damage was negIigible (i.e., in a linear viscoelastic
Pseudo Strain
range). Therefore, from the study of monotonie loading data,
e::
the value of 200 was assumed for both mixtures as an upper FIG. 8. CyclIc Behavlor ot A.phalt Concrete under Controlled-
Stre•• Mode: (a) (J' - E; (b) (J' - ER
limit of this linear viscoelastic range. The regression analysis
on the data shown in the figure resulted in the following fonn:
With the use of the fint derivatives of these strain functions

=G [~o + ~~]
and the relaxation modulus in the fonn of pure power law
G2 (eg) 2R (18) (i.e., E(t) = Elt- m ), the following pseudo strain can be ob-
tained:
The regression coefficients in (18) are summarized in Table 4
(19)
for the two mixtures.
where Po and PI = constants depending on Bh Eh m, a, €O,
Mode of Loadlng Independent Constltutive Model and w. The ef can be obtained by subtracting the amplitude
of pseudo strain from the average pseudo strain of each cycle
Before the controlled-strain constitutive equation [see (15)] (i.e., pot",-m) - PI)' For a linear viscoelastic case, a = m and
is applied to the controlled-stress mode, a careful investigation Po = Pl because the pseudo strain is the same as the stress
was made on the controlled-stress fatigue test data to identify (i.e., O' = 0'0 + 0'0 sin(wt + <\>')) based on the definition of
the basic difference in hysteretic behavior and damage growth pseudo strain in (1). Thus, ef = O, as seen in Fig. 3(b). When
under both modes of loading. Fig. 8 present typical stress- the load level is high enou~h to induce some damage to a
strain and stress-pseudo strain relationships. It can be seen specimen, a > m resulting es > O, as shown in Fig. 8(b).
from Fig. 8(b) that the two characteristics of the stres s-pseudo Considering the foregoing observations and increasing de-
strain behavior observed from the controlled-strain mode (i.e., fonnation in the controlled-stress test in tension, it can be said
the hysteresis loops and the reduction in the secant pseudo that the presence of ef is caused by the pennanent defonnation
stiffness as cyc1ic loading continues) appear to be valid in the accumulated in the specimen during the tensile fatigue test.
controlled-stress mode also. However, in the controlled-stress Therefore, it is proposed that ef be used as the last ISV in the
mode, an additional stres s-pseudo strain characteristic needs to constitutive modeling.
be considered. That is, the hysteresis loop shifts away from Using the controlled-stress fatigue test data, the Sp values
the origin as cyclic loading continues. This shift in the pseudo were calculated from (5) with p = 2(1 + 11m). Different from
strain is denoted by ef in Fig. 8(b). the controlled-strain case, Sp was not able to eliminate the
To understand the characteristics of ef, it is useful to study stress-level dependency. Therefore, p =21m was attempted and
the analytical solution of pseudo strain for the strain history successfully handled the stress-level dependency as shown in
under the controlled-stress mode given in (9) and (10). Be- Fig. 9(a). These observations suggest that k = (1 + 11m) is a
cause the first tenn Bo and the last tenn on the right side of better assumption for the controlled-strain mode whereas k =
(9) have little effect on the average strain in each cycle, el(t) 11m is adequate for the controlled-stress case. Based on the
is simplified to Bit"'. Also, a constant strain ampIitude €o is theoretical relationship between the failure zone characteristics
assumed instead of €o(t) in (lO) for the simplicity of analysis. and k value discussed in the section Damage Parameter, it is
JOURNAL OF ENGINEERING MECHANICS / JANUARY 1998/37
 2.0 r-------------,------n reduction in the initial secant pseudo stiffness to eliminate the
• viscoelasticity of the materials. Then, the value of SI is the one
.OOO6unit
~ • corresponding to the 50% of the initial secant pseudo stiffness
~ 1.5 .0014unit as shown in Fig. 9(a). In the controlled-strain mode, the SI
= Controlled-Slress c values were 1,900 and 2,100 for AAD and AAM mixtures,
~ 32 kPa respectively, and in the eontrolled-stress mode, 760 and 2,000
O o
-g 1.0 115kPa
for AAD and AAM mixtures, respectively. The normalized
CI)
II)
---~---~___ damage parameter Sn is defined as follows:
a.. F(Sp}
E S _ Sp
(20)
~ 0.5 I - - - - - , , - - - - - - - - I I I K n - SI
CI) 50% ol initial S A
(a) The Sp in (16) is then replaced with Sn
0.0 L -_ _' - -_ _' - -_ _ '--....:...!:~'--_---l

o 500 1000 1500 2000 2500 (21)

Sp The ~ values shown in Fig. 9(a) were plotted against the
Sn values in Fig. 9(b). As can be seen in this figure, all the SR
2.0 r------------..,....--"'"I1 values for both modes of loading falI on the same eurve.
• To take advantage of the constitutive relationship developed
.OOO6unit from the controlled-strain tests, the constitutive equation [see
~ • (15)] is extended to account for the controlled-stress ease.
~ 1.5 .0014unit
First, to move the stress-pseudo strain eurves in Fig. 8(b) baek
= c
to the origin, the eR term in (15) is replaced with (eR - e:) as
~ 32 kPa
o o follows:
-g 1.0
5l
a.. (22)
E
~ 0.5 where t R = eR - e:.
CI)
CI) Because ~ is not able to represent the slope of each cycle
0.0 L-_.....J.._ _-'--_---'_ _........._ _'--_-J
(b) because of e:,
the slope of each eycle is denoted here by §R
as sho-.yn in the figure. Using a geometrical relationship be-
O 0.2 0.4 0.6 0.8 1.2 tween SR and SR, the following relationship can be established:
Sn
FIG. 9. Reductlon In SR under Both Mode. of Loadlng (AAD
mlxture): (a) SR_S,,; (b) SR_S.
F(Sn) = (1 + Efef) F(Sn) (23)

where fr = 8R; F = ~; and tf = ef - e:. It is noted here that

possible to speeulate that the material's fracture energy (r) and the coefficients for F(Sn) in (23) can be obtained from the
failure stress are eonstant under the eontrolled-strain mode controlled-strain data because the validity of F(S.) for both
while the fraeture proeess zone size and r are eonstant under modes of loading has been verified earlier. Replacing F(Sn) in
the eontrolled-stress mode. (22) with (23), the final form of the constitutive model is ob-
Beeause the reduetion in ~ was observed from both modes tained

(1 + :D F(S.) + G(eg, :;) ]

of loading, the eharaeteristies of ~-Sp relationship were eom-
pared in Fig. 9(a). It is observed from this figure that the Sp er = [(ER) [ (24)
value of the eontrolled-strain mode is larger than the one of
the eontrolled-stress mode for a given value of ~. This dis- where the function F and G are given in (21) and (17), re-
erepaney may be attributed to several faetors. In the eon- spectively.
trolled-strain fatigue tests, both tensile (a positive sign) and
eompressive (a negative sign) stresses oeeur in a specimen,
In the controlIed-strain mode, e:
= O; thus (24) is identical
to (15). Furthermore, this eonstitutive equation [see (24)] can
whereas only tensile stresses oceur in the speeimen under the be simplified for monotonie loading as follows:
eontrolled-stress mode. Schapery (1981) proposed the damage
parameter in terms of leRI instead of eR to include some dam- (25)
age because of the eompressive stresses. However, most of the beeause e: = O and funetion G .... O when eR = ef. A simi1ar
cracks observed during the controlled-strain fatigue tests con- result has been observed by Kim and Little (1990). When dam-
dueted in this study were tensile cracks, which propagated in age is smalI (i.e., time elapsed is smalI), it can be seen from
the direction perpendicular to the loading direction. This im- (21) that F(Sn) ~ 1 because Sn ~ O. As a result, (25) is ide n-
plies that the compressive stresses caused little damage to the tical to the linear viscoelastic constitutive equation [see (2)]
specimen. Furthermore, the compressive stresses have benefi- because ER = 1. This again verifies the linear behavior shown
cial effects on the closing of tensile cracks. Therefore, the use in Figs. 3(b) and 4(b) for negligible damage.
of leRI in the eontrolled-strain mode may result in the over-
estimation of the damage. VERIFICATION OF CONSTITUTIVE MODEL
For the foregoing reasons, Sp was normalized by using a
value of Sp at failure, denoted by SI. To determine the value For the verification purpose, the folIowing two types of uni-
of SI' the failure eriterion of a specimen should be established axial testing were performed at 25°C:
a priori. The most common failure criterion of asphalt concrete
under the eontrolled-strain mode is a 50% reduction in the 1. Controlled-strain fatigue tests with the strain amplitudes
initial stiffness (Yoder and Witczak 1975). Knowing the stiff- of 0.0009 and 0.001 units for AAD and AAM mixtures,
ness reduetion is eaused by both relaxation and damage growth respeetively
of the materials, this failure eriterion is modified to the 50% 2. ControlIed-stress fatigue tests with the stress amplitudes
38/ JOURNAL OF ENGINEERING MECHANICS / JANUARY 1998
 of 58 and 129 kPa for AAD and AAM mixtures, respee- fatigue test is not as good as the prediction in the controlled-
tively strain ease. It must be noted here that the coefficients in (24)
used in the predietion were obtained solely from the con-
In addition to the fatigue test data, the monotonie loading data trolled-strain fatigue tests. The extrapolation from the con-
used in the experimental study of the eorrespondenee prineiple trolled-strain mode to the eontrolled-stress mode might have
were used in the verifieation study. affeeted the accuracy. Another reason for the reduced accuraey
Fig. 10 (AAD mixture with N, = 44,000) demonstrates the is attributed to some errors in the calculation of pseudo strain
verifieation of the model for the eontrolled-strain fatigue tests, because of poor resolution of the measured strain. In the con-
Fig. 11 (AAM mixture N, = 39,000) for the eontrolled-stress trolled-stress fatigue tests, the deformation of a specimen is
fatigue tests, and Fig. 4(a) for the monotonie tests. To validate continuously increasing as cyclie loading continues; thus the
the constitutive model for entire fatigue life of asphalt con- capacity of the deformation measuring deviee should be large
erete, four cycles were selected randomly for presentation, enough to cover the entire loading history. This results in the
which represent initial, early, intermediate, and final states of poor resolution for deformation response of the individual ey-
fatigue life of the specimens. In general, an excellent agree- cles. Although the prediction within each eyele is not as good
ment is found between the measured and predicted stresses. as the eontrolled-strain ease, the eonstitutive model reasonably
In Fig. lO, the constitutive model successfully predicts the prediets the stresses up to failure, indieating the applieability
reduction in stresses in the controlled-strain test all the way of the eonstitutive model to different modes of loading.
up to failure. Because the strain amplitude used in the verifi- The eonstant-strain-rate monotonie test results are presented
eation test was not used in the determination of the eoefficients in Fig. 4(a) for AAD mixtures. Test results are presented all
in the eonstitutive equation [see (24»), it is proved again that the way up to the ultimate stress level. Overall, the predietion
Sp is an excellent means of eliminating the strain-Ievel depen- is in a good agreement with the experimental data for the
denee of the materials. entire loading history regardless of the strain rates, demon-
In Fig. 11, the accuraey of prediction in the controlled-stress strating the rate independenee of the proposed model. It is
noted also that the eonstitutive model developed from the cy-
clic loading suecessfully prediets the stress-strain relationship
800 under the monotonie loading.

CONCLUSIONS
Damage aeeumulation under uniaxial tensile eyclie loading
(bl without rest periods was modeled using the elastie-viseoelastie
(al
·600 '------~--'--' ~~-~--~-~ eorrespondenee principle and the time-dependent damage pa-
O 0.0004 0.0008 0.0012 O 0.0004 0.0008 0.0012
Straln
rameter. The pseudo variabies greatly simplified the task of
Strain
separately evaluating viseoelastie behavior and damage growth
~or---~-----, ~0r-----~-.3-aa-h-~ of asphalt eonerete. The damage parameter sueeessfully elim-
9.589Ih cycIe
inated stress-strain-Ievel dependence of asphalt conerete on
800 800
\i' fatigue behavior. The eonstitutive equation developed from the
~300 eontrolled-strain fatigue tests satisfaetorily predieted the stress-
j O~~--~~~--~ strain behavior of asphalt eonerete all the way up to failure
under the eontrolled-stress mode as well as under the mono-
-300 tonie loading with varying strain rates.
(el (dl
~O'------~--'--' ·600 '---~--~----
o 0.0004 0.0008 0.0012 o 0.0004 0.0008 0.0012 ACKNOWLEDGMENTS
SlrIln Strain
The experimental data used in this paper were generated from a re-
FIG. 10. Valldatłon ot Constltutlve Model under Controlled- search project sponsored by Texas A&M Research Foundation, Western
Straln Mode wlth Straln Amplltude ot 0.0009 unit Research Institute, and Federal Highway Adrninistration. The writers are
grateful for these organizations' support.
2OOr-=~------, 200,...-----------,
lat cycIe 153rdcycle
APPENDIX I. REFERENCES
Kim, Y. R, and Little, D. N. (1990). "One-dimensional constitutive mod-
eling of asphalt concrete." J. Engrg. Mech., ASCE 116(4), 751-772.
Kim, Y. R, Lee, Y., and Lee, H. J. (1995). "Correspondence principle
for characterization of asphalt concrete. " J. Mat. in Civ. Engrg., ASCE,
(al 7(1), 59-68.
(bl
O~-~-----~ Reddy, J. N., and Rasmussen, M. L. (1982). Advanced engineering anal-
0.0001 0.0002 0.0006 0.0008 0.001 ysis. John Wiley & Sons, Inc., New York, N.Y.
SIrain Strain Schapery, R A. (1981). "On viscoelastic deformation and failure behav-
ior of composite materials with distributed flaws." 1981 Advances in
2OOr--------,
10,o48IIIh cycIe
2OOr---------......,
34,G01h cycIe aerospace structures and materia/s, S. S. Wang and W. J. Renton, eds.,
American Society of Mechanical Engineers, New York, N.Y., 5-20.
Schapery, R A. (1984). "Correspondence principles and a generalized J-
integraI for large deformation and fracture analysis of viscoelastic me-
dia." lnt. J. Fracture, 25(1), 95-223.
Yoder, E. J., and Witczak, M. W. (1975). Principles ol pavement design.
John Wiley & Sons, Inc., New York, N.Y.
(el (dl
O'----~-----~ o~---~~------ APPENDIX II. NOTATlON
0.0093 0.009S 0.0097 0.0093 0.00II5 0.00117
Strain Strain The following symbols are used in this paper:
FIG. 11. Valldatlon of Constltutlve Model under Controlled-
Stresa Mode wlth Stress Amplltude ot 129 kPa D(t) = uniaxial creep compliance;
JOURNAL OF ENGINEERING MECHANICS / JANUARY 1998/39
E(t) = uniaxial relaxation modulus; !r = slope of stress-pseudo strain curve in the controlled-stress
ER = reference modulus; mode;
k = exponent in crack speed-generalized J-integral relation- t = time elapsed from the first application of loading;
ship; ex = exponent for average strain-time relationship under the
m = exponent in creep compliance-time relationship in the controlled-stress mode;
pure power law; e = uniaxial strain;
N = exponent in stress-absolute pseudo strain relationship; eR = uniaxial pseudo strain;
p = exponent in damage parameter; ef = largest value of pseudo strain up to the current time;
SI = damage parameter at failure; e~ = amplitude of pseudo strain;
S,. = internal state variabies; e: = starting value of stress-pseudo strain cycle in the con-
S. = normalized damage parameter; trolled-stress mode;
S~ = damage parameter; (J' = uniaxial strain; and
S = secant pseudo stiffness; T = time variable of integration.

40 I JOURNAL OF ENGINEERING MECHANICS I JANUARY 1998