You are on page 1of 26

Backcalculation of Dynamic Modulus from Resilient Modulus Test Data

For the past few decades, the stiffness of the material used for roadway design and construction has
been commonly characterized by means of resilient modulus. However, the resilient modulus is not a
fundamental material property, and hence, the concept of resilient modulus has been subsequently
diminished in the latest Mechanistic-Empirical Pavement Design Guide (MEPDG). Although the
MEPDG could not endorse the use of the resilient modulus test protocol as the primary means of
asphalt concrete modulus characterization, it has been a primary mixture test, and a considerable
amount of laboratory testing has been done to date. In this paper, an analysis methodology for
backcalculating the dynamic modulus from the resilient modulus test data is introduced. The strong
advantage of such a methodology is that the existing resilient modulus data can be reused for
estimating the dynamic modulus. The approach would significantly save time and effort in re-
evaluating the dynamic modulus of asphalt mixture for which the resilient modulus test data are
available.

Keywords: Asphalt Mixtures, Viscoelasticity, Creep Compliance, Resilient Modulus, Dynamic Modulus

1
1. INTRODUCTION

Unlike an elastic material, the fundamental properties that govern the stress-strain relationship of

a viscoelastic material are functions of either time or frequency. In the time domain, these

properties are defined as the response of the material subject to either a unit step stress (creep

compliance) or a unit step strain (relaxation modulus). On the other hand, the fundamental

properties in the frequency domain, such as the complex modulus and compliance, are defined as

the response of the material subjected to sinusoidal loading at a given frequency. According to

the theory of linear viscoelasticity, the response of a viscoelastic material subjected to any given

stress or strain loading history can be predicted through the well-known convolution integral

provided that the appropriate fundamental property is known a priori. In addition, the above-

mentioned fundamental properties are interchangeable, meaning that if one is known, then the

other can be obtained through the available interconversion methods (Findley et al. 1976,

Tschoegl 1989, Wineman and Rajagopal 2000, Park and Schapery 1999, Schapery and Park

1999). Therefore, the theoretical response of a viscoelastic material loaded to any given

arbitrary excitation can be predicted if one of the aforementioned fundamental properties is

known. Such an argument would also imply that the fundamental properties can be

“backcalculated” from a pair of arbitrary loads and their corresponding response because the

response of the material under the arbitrary loading condition should also be unique.

For the past few decades, the stiffness of asphalt concrete mixtures used for roadway

design and construction has been commonly characterized by means of resilient modulus defined

as the ratio of the applied stress to the recoverable strain (AASHTO 1993). The loading

function for this test typically consists of a 0.1-second haversine curve followed by a 0.9-second

2
rest period. However, the resilient modulus is not a fundamental material property, and hence,

the concept of resilient modulus has been subsequently diminished in the latest Mechanistic-

Empirical Pavement Design Guide (ARA 2004). Instead, the new design guide uses the

dynamic modulus, which is the magnitude of the complex modulus, as the primary measure of

stress-strain behavior of the asphalt mixtures in the procedure for both new construction and

rehabilitation projects. The design guide recommends that the dynamic modulus of asphalt

mixtures be estimated based on the laboratory test results using the dynamic modulus test

specified in AASHTO TP 62.

In this paper, an analysis methodology is presented for determining the fundamental

property of a viscoelastic material from an arbitrary function by means of the method of least

squares. The concept of the backcalculation approach is that if the historical resilient modulus

test data are available for the given asphalt mixture, then it is possible to backcalculate the

dynamic modulus from the test data by treating the resilient modulus loading as the arbitrary

loading. The paper begins with a presentation of a numerical method for evaluating the

convolution integral, followed by a least squares backcalculation methodology for determining

the creep compliance of a viscoelastic material in a uniaxial loading state. Then, the analysis

methodologies are extended and applied to the biaxial loading test, namely the indirect tension

(IDT) test (Buttlar and Roque 1994, Zhang et al. 1997, Kim et al. 2004, and Kim et al. 2005).

The creep compliances backcalculated from the uniaxial and the IDT resilient modulus tests are

then converted into the dynamic moduli by use of the Prony series. The backcalculated results

are then compared to the actual uniaxial and IDT dynamic modulus test results.

3
2. DEVELOPMENT OF A BACKCALCULATION METHODOLOGY FOR THE
UNIAXIAL TESTING CONDITIONS

2.1. Numerical Evaluation of the Convolution Integral

Based on the theory of linear viscoelasticity, the relationship between the time-dependent

uniaxial stress and strain (Figure 1) can be expressed as a convolution integral of the following

form.

¥
¶s (t )
e (t ) = ò D(t - t ) dt (1)

¶t

where e(t) is strain, D(t) is creep compliance, and s(t) is stress.

[INSERT FIGURE 1 HERE]

The above convolution equation, however, can be also written as (Wineman and

Rajagopal, 2000):

¥

e (t ) = ò s (t ) D(t - t )dt

¶ (t - t )
¥
(2)

= - ò s (t ) D(t - t )dt

¶t

Without any loss of generalization, the two functions D(t) and s(t) can be expressed as:

D(t ) = D(t )H (t ) (3)

s (t ) = s (t )H (t ) (4)

where H(t) is the Heaviside step function. Taking the derivative of Equation (3) yields:

¶ ¶
D(t ) = D(t ) × d (t ) + H (t ) × D(t ) (5)
¶t ¶t

where d(t) is the diarc delta function. Also by means of the chain rule, the derivative of D(t-t)

with respect to t in Equation (2) can be obtained as:

4
¶ ¶ ¶D(t ¢)
D(t - t ) = - D(t - t ) = - (6)
¶t ¶ (t - t ) ¶t ¢ t ¢=t -t

Substituting Equations (4), (5), and (6) into Equation (2) yields:

¥ ¥

e (t ) = ò s (t )H (t ) D(t - t )d (t - t )dt - ò s (t )H (t ) H (t - t ) ¶t D(t - t )dt (7)
-¥ -¥

The variables inside the integral can be defined for the range t > t ³ 0 - . Using the properties of

H(t) and d(t), the above equation can be simplified further and results in the following equation,

known as the Riemann-Stieltjes integral (Wineman and Rajagopal, 2000):

t

e (t ) = D(0) × s (t ) - ò s (t ) D(t - t )dt (8)
0
¶t

In order to evaluate the above equation numerically, the integral needs to be evaluated for each

given time step, tn, as follows:

tn

e (t n ) = D(0) × s (t n ) - ò s (t ) D(t n - t )dt
0
¶t
t2 3 t
¶ ¶
= D(0 ) × s (t n ) - ò s (t ) D(t n - t )dt - ò s (t ) D(t n - t )dt K
t1 = 0
¶t t2
¶t
tn (9)

- ò s (t ) D(t n - t )dt
t n -1
¶t
ìïti +1
n -1
¶ üï
= D(0 ) × s (t n ) - å í ò s (t ) D(t n - t )dt ý, for t n = t1 K t N
i =1 ïî ti ¶t ïþ

Since s(t) = si for any discrete time interval ti < t < ti+1, it can be taken out of the integral, and

the above equation can be expressed as shown below.

n -1 ì ti +1 üï
ï ¶
e (t n ) = D(0) × s n - å ís i × ò D(t n - t )dt ý, for t n = t1 K t N (10)
i =1 ï ¶t
î ti þï

5
In Equation (10), the only function remaining inside the integral is the derivative of the creep

compliance; therefore, it can be evaluated easily. Evaluation of the integral in a closed form

yields:

ti +1

ò D(t n - t )dt = D(t n - t i +1 ) - D(t n - t i ) (11)
ti
¶t

The equation for the numerical evaluation of the convolution is finally obtained by substituting

the above into Equation (10):

n -1
e (t n ) = D(0) × s n + å [s i × {D(t n - t i ) - D(t n - t i +1 )}], for t n = t1 Kt N (12)
i =1

Note that the above equation can be applied to any arbitrary functions D(t) and s(t) that are

available numerically.

2.2. Backcalculation of Creep Compliance Using Least Squares

In order to backcalculate the creep compliance using the least squares approach, it is necessary to

identify a function that represents the creep compliance of viscoelastic materials. Because of the

nature of least squares, it is also important that the function have a simple form with the least

number of coefficients to minimize the computation time and effort. The power function, which

has been well accepted as an analytical representation of the creep compliance of viscoelastic

materials (Findley et al. 1976, Roque et al. 1997, Kim et al. 2005, and Kim et al. 2008), appears

to be the most promising and simplest function for the purpose. The power function is expressed

as follows:

D(t ) = D0 + D1 × t m (13)

Then, the following equation is obtained by substituting Equation (13) into Equation (12):

6
n -1
[ { }]
e (t n ) = D0 × s n + D1 × å s i × (t n - t i )m - (t n - t i +1 )m , for t n = t1 Kt N (14)
i =1

which can be used to calculate the time-dependent strain of a viscoelastic material with the D(t)

shown in Equation (13) subjected to any arbitrary loading s(t).

In order to determine the power function parameters D0, D1, and m using the least squares

approach from the strain output corresponding to any arbitrary stress input, the objective function

is set up as follows:

N
F = å {e (t n ) - e m,n }2 (15)
n =1

where F is the function to be minimized, and em,n is the measured strain at time t = tn. Upon

substituting Equation (14) into the above, one finds:

N
F = å {D0s n + D1x n - e m,n }2 (16)
n =1

where

n -1
[ {
x n = å s i × (t n - t i )m - (t n - t i +1 )m }] (17)
i =1

It should be noted that the objective function shown in Equation (16) leads to a non-linear least

squares problem because of the parameter m in Equation (17). However, if a direct search

method (that is, the m value is known prior to performing each iteration.) is used for the m value,

then the problem can be simplified to a linear least squares problem in which the only unknowns

in Equation (16) are the remaining parameters of the power function, D0 and D1. Therefore, in

order for the F function to be minimized for a given m value, its partial derivative with respect to

D0 and D1 should be equal to zero, as shown below.

7
N
¶F
= 2 × å {D0s n + D1x n - e m,n }×s n = 0 (18-1)
¶D0 n =1

N
¶F
= 2 × å {D0s n + D1x n - e m,n }× x n = 0 (18-2)
¶D1 n =1

In the matrix form, the solution of the above set of linear equations can be expressed as:

{D} = [A]-1 {B} (19)

where {D} = {D0 D1 }T , and the matrix [A] and the vector {B} are shown below.

é N 2
N ù
ê ås n å s nx n ú
[A] = ê Nn=1 n =1
N
ú (20)
ê 2 ú
êå s n x n åxn ú
ë n =1 n =1 û

T
ìN N ü
{B} = íå e m,ns n å e m,nx n ý (21)
î n =1 n =1 þ

2.3. Conversion of Creep Compliance to Dynamic Modulus

Complex modulus, which is a fundamental property of a viscoelastic material defined in the

frequency domain, can be determined from sinusoidal loading applied at different frequencies.

For a viscoelastic material, if the input is an oscillatory stress, s0eiwt, where w is the angular

frequency, then the strain response, e0eiwt, will be an oscillation at the same frequency as the

stress but lagging behind by a phase angle, d. From the complex modulus test, complex E*,

dynamic |E*|, storage E¢, and loss E¢¢ moduli can be determined as follows:

s 0 id s 0
E* = e = (cos(d ) + i sin(d )) = E * × (cos(d ) + i sin(d )) = E '+iE" (22)
e0 e0

8
Similar to the complex modulus, complex compliance, D*, which can be determined from an

oscillatory strain input, is expressed as follows:

e 0 - id e 0
D* = e = (cos(d ) - i sin(d )) = D * × (cos(d ) - i sin(d )) = D'-iD" (23)
s0 s0

Since D* and E* are reciprocals of each other, the dynamic |D*|, storage D¢, and loss D¢¢

compliances can also be determined from the complex modulus test. These fundamental

properties determined in the frequency domain are interconnected with those determined in the

time domain, such as creep compliance D(t) and relaxation modulus E(t), by the following

relationships (Findley et al. 1976):

E * (w ) =iw × L | E (t ) | s =iw (24)

D * (w ) =iw × L | D(t ) | s =iw (25)

) ¥
where L[ f (t )] = f (s ) = ò e - st f (t )dt is the Laplace transform of the function f (t ) , and s is the
0

Laplace variable.

Although the power function has been successfully used for representing the creep

behavior of viscoelastic materials, its mathematical deficiency does not provide enough

flexibility for analytical interconversion of the fundamental properties of viscoelastic media.

Instead, Prony series has been widely used for the purpose of interconversion because of its

mathematical efficiency. For creep compliance in the time domain, the Prony series

representation is of the following form:

t
N -
ti
D(t ) = D0 + å Di (1 - e ) (26)
i =1

where D0, and Di, are the Prony series parameters, and ti are the retardation times. In fitting

creep compliances using the Prony series, the retardation times, ti, are usually specified, and then

9
the unknown coefficients are determined by solving the linear system of equations (Park and

Kim, 2001 and Kim et al., 2008). In this study, Prony series shown in Equation (26) was fitted

for the backcalculated creep compliance, expressed in terms of the power function, with N=7 and

retardation times of one decade interval, ti = 10(i-5) (i = 1,¼ 7).

Once the Prony series parameters have been determined, the real and imaginary parts of

the complex compliance can be obtained by substituting Equation (26) into Equation (25):

N
Di
D' (w ) = D0 + å (27)
i =1 w ti2 +1
2

N
wt i Di
D' ' (w ) = å (28)
i =1 w 2t i 2 + 1

Finally, the dynamic modulus is obtained as the reciprocal of the complex compliance as shown

in the equation below:

1 1
E * (w ) = = (29)
D * (w ) D' (w ) + D' ' (w ) 2
2

3. EXTENSION OF THE BACKCALCULATION METHODOLOGY TO THE


INDIRECT TENSION TESTING CONDITIONS

3.1. Elastic Stress Analysis for Indirect Tensile Test

The illustration of the IDT test and the plane-stress distribution within an IDT sample are shown

schematically in Figure 2.

[INSERT FIGURE 2 HERE]

For an IDT specimen with radius R and thickness d subjected to a strip load of width a =

2R·sin(a) and magnitude P, the distributions of the tensile stress along the horizontal axis and the

10
compressive stress along the vertical axis were given by Hondros (1959) under the assumption of

plane-stress conditions as:

2 2
2P é 1 - ( x R ) sin 2a -1 1 - ( x R )
æ öù
s x ( x, 0 ) = ê 2 4
- tan ç
ç 1 + (x R ) 2
tan a ÷ú
÷
pad êë1 + 2( x R ) cos 2a + ( x R ) è øúû (30-1)
2P
= m( x )
pad

2 2
2P é 1 - ( y R ) sin 2a -1 1 + ( y R )
æ öù
s y (0, y ) = - ê 2 4
+ tan ç
ç 1 - ( y R )2 tan a ÷ú
÷
pad êë1 - 2( y R ) cos 2a + ( y R ) è øúû (30-2)
2P
=- n( y )
pad

from which the average stresses over the gauge length are calculated as (Lee and Kim 2009 and

Kim and West 2010):

2P 1 l 2
s x ,avg = × m( x )dx (31-1)
pad l ò- l 2

2P 1 l 2
s y ,avg = - × n( y )dy (31-2)
pad l ò- l 2

Because of the nature of the IDT specimen geometry, another constant can be derived. This

constant, b, is defined as the ratio of the horizontal and vertical stress magnitudes and will be

used later to simplify the equations for the creep compliance backcalculation:

l
ò- l 2 m(x )dx
2
s x ,avg
b= = l
(32)
s y ,avg
ò
2
-l
n ( y )dy
2

With the above definition for b , the relationship between the two stresses can be written as:

s y , avg = - b × s x , avg (33)

For simplicity, the average horizontal and vertical stresses will be designated respectively as sx

and sy in the subsequent sections of the paper.

11
3.2. Determining the Creep Compliance from Random Loading in the IDT Test

The plane-stress constitutive equations for a linear, elastic, and isotropic material are given as:

e x = D × [s x - ms y ] (34-1)

e y = D × [s y - ms x ] (34-2)

where D and m are the elastic compliance and Poisson’s ratio for the given material. According

to the elastic-viscoelastic correspondence principle, the plane-stress constitutive equations for a

viscoelastic material in the Laplace domain are given as:


) ) ) ) )
e x ( s ) = s D(s ) × [s x (s ) - sm (s )s y (s )] (35-1)

) ) ) ) )
e y ( s ) = s D(s ) × [s y (s ) - sm (s )s x (s )] (35-2)

)
Rearranging the terms after eliminating m (s ) from Equations (35-1) and (35-2) results in the

following:
) )
) s y (s ) ) ) ì) s y (s ) ) ü
e x (s) - ) e y ( s ) = s D(s )ís x (s ) - ) s y (s )ý (36)
s x (s ) î s x (s ) þ

The above equation can be further simplified by use of b defined in Equation (32) as:
) ) ) ) )
e x ( s ) + be y ( s ) = s D(s ) × [s x (s ) + bs y (s )] (37)

Taking the inverse Laplace transform of the above equation yields:

t
{e (t ) + be (t )} = ò D(t - t ) ¶ {s (t ) + bs (t )}dt
x y x y (38)
0-
¶t

Note that the above equation is essentially identical to Equation (1) if the following substitutions

are made for the stress, s(t), and strain, e(t), terms in Equation (1):

e (t ) = {e x (t ) + be y (t )} (39)

12
s (t ) = {s x (t ) + bs y (t )} (40)

Then, with the above replacement of variables and the power function shown in Equation (13)

for creep compliance, the same derivation and numerical scheme outlined for the uniaxial case

[Equations (2) through (18)] can be followed in backcalculating the creep compliance for the

biaxial loading conditions. By using the direct search method for the m value, the remaining

power function parameters D0 and D1 can be determined from Equation (19), with the following

expressions for matrix [A] and vector {B}.

é N N ù
(
ê å s x ,n + bs y ,n
2
) å (s x,n + bs y ,n )x n ú
[A] = ê nN=1 n =1
N
ú (41)
ê ú
(
êå s x ,n + bs y ,n x n ) åxn 2
ú
ë n =1 n =1 û

T
ìN N ü
( )(
{B} = íå e x,n + be y ,n s x,n + bs y ,n ) å (e x,n + be y,n ) xn ý (42)
î n =1 n =1 þ

where

n -1
[ {
x n = å (s x ,i + bs y ,i ) × (t n - t i )m - (t n - t i +1 )m }] (43)
i =1

In subsequent sections, the results of the proposed backcalculation methodology evaluated for

laboratory experiments using both the laboratory-made and field-cored specimens will be

presented.

4. EXPERIMENTAL STUDY USING LABORATORY-MADE MIXTURES

For the laboratory-made asphalt mixtures, two testing modes—the uniaxial compressive test and

the indirect tensions (IDT) test—were used for evaluating the proposed backcalculation

13
methodology. The former, which is the primary testing mode for the MEPGD, is normally

conducted in the intermediate to high temperature range (approximately 0°C to 60°C), while the

latter, which has been used for evaluating the resilient modulus, creep, and tensile strength

(Roque and Buttlar, 1992, Buttlar, and Roque, 1994, and Roque et al., 1997) of asphalt mixtures,

is conducted in the low to intermediate temperature range (approximately -30°C to 30°C). It was

therefore decided to evaluate the backcalculation methodology for dynamic modulus in the

intermediate temperature range at 10ºC.

4.1. Specimen Preparation

The material selected for testing was a densely graded asphalt mixture meeting Superpave

requirements designed and provided by the Florida Department of Transportation (FDOT). The

mixture had a nominal aggregate size of 12.5 mm and an unmodified binder graded as PG 67-22.

This is the primary mixture type used by FDOT

For uniaxial compression testing, two replicate specimens 170 mm tall by 150 mm in

diameter were prepared with a Superpave Gyratory Compactor. These gyratory specimens were

cored and trimmed to a final size of 150 mm tall by 100 mm in diameter in accordance with

AASHTO TP 62. The air void content of the cored and trimmed specimens was 4 percent, with a

tolerance of ± 0.5 percent. For IDT testing, three replicate specimens 115 mm tall by 150 mm in

diameter were prepared with the same Superpave Gyratory Compactor. Those were carefully cut

to a thickness of 38 mm. The air void content of the specimens was 7 percent, with a tolerance of

± 0.5 percent in accordance with AASHTO T 322.

4.2. Testing on Laboratory-Fabricated Mixtures

14
A picture of the uniaxial test setup with specimen dimensions is shown in Figure 3. For the

uniaxial test, dynamic modulus and resilient modulus tests were performed on the prepared

specimens. The resilient modulus test was conducted for five cycles in which each cycle

consisted of a 0.1-second haversine load followed by a 0.9-second rest period. Although it is a

rough estimate, the time range covers the frequency range from 0.03 Hz to infinity based on the

relationship proposed by Christensen and Anderson (1992). The data were measured according

to the method documented in a previous study (Roque et al. 1997), and the same method was

applied to the IDT resilient modulus test. Then, dynamic modulus tests were conducted at

various frequencies—0.1 Hz, 0.5 Hz, 1.0 Hz, 5.0 Hz, 10.0 Hz, and 25.0 Hz—with rest periods of

at least four minutes between the individual tests.

[INSERT FIGURE 3 HERE]

After the uniaxial test, the IDT resilient modulus test was conducted at the same

temperature and for the same number of loading cycles. A picture of the IDT testing setup is

shown in Figure 4. The specimen geometry and the system for measuring the stress and strain

responses were in accordance with those recommended in AASHTO T 322. The maximum strain

in the resilient modulus test performed under the uniaxial testing mode and the maximum

horizontal strain in the same test under the IDT testing mode were kept below 100 me.

[INSERT FIGURE 4 HERE]

4.3. Verification of the Numerical Approach

Once the creep compliance by means of the power function was backcalculated from the

arbitrary loading in the resilient modulus test under the uniaxial testing mode, Prony series was

then fitted to the power function as described above to determine the dynamic moduli at multiple

15
frequencies. In order to verify the accuracy and reliability of the backcalculation methodology,

the backcalculated uniaxial creep compliance represented in both the power function and in

Prony series was reused as input into the numerical convolution integral (Equation 12). If the

predicted strain from the backcalculated creep compliance is well fitted to the measured strain,

then the error between the predicted and measured strains should be minimal, and the approach

can be reliably extended to the biaxial testing mode, that is, the IDT test. Figure 5 shows the

measured strains compared to those predicted. As seen in the figure, both the power function and

Prony series show excellent agreement with the measured strains. This ensures that the power

function parameters as well as the Prony series parameters can be reliably backcalculated from

the resilient modulus data.

[INSERT FIGURE 5 HERE]

4.4. Comparison of Dynamic Modulus from Backcalculation to Measured Dynamic Modulus

Because the backcalculated uniaxial creep compliance was successfully used to predict the

strains from the resilient modulus test, the dynamic modulus was estimated from the creep

compliance backcalculated from both the uniaxial and IDT resilient modulus data. This

interconversion was carried out through Prony series as was shown above for both the uniaxial

and IDT test results. Figure 6 shows a plot between the dynamic moduli backcalculated from the

uniaxial and the IDT resilient modulus tests against those measured from the uniaxial complex

modulus test at 10ºC. For the sake of clarity, the dynamic moduli shown in the figure are

averages of the replicate specimens. The backcalculated moduli from the uniaxial resilient

modulus test were slightly higher than those measured from the uniaxial dynamic modulus test.

The average difference of the backcalculated dynamic modulus over the measured dynamic

16
modulus was 7 percent. Considering that the variability of 10 percent is normally accepted as a

tolerance of mixture tests, the dynamic modulus from the uniaxial resilient modulus test appears

to be in the acceptable range of mixture test results. However, the dynamic modulus values

backcalculated from the IDT resilient modulus test were lower than those measured in the

uniaxial test: the difference was 23 percent. Given that the air voids of 7 percent were designed

for the IDT specimen, which were 3 percent higher than those of the uniaxial testing specimens,

the lower modulus values were expected. Based on the known volumetric properties of the given

mixture (Table 1), a reduction of 15 percent in dynamic modulus could be anticipated from the

well-known predictive equation (Witczak et al. 2002). This indicates that the dynamic modulus

values would have been approximately 15 percent lower than those measured in the uniaxial

dynamic modulus test if a uniaxial specimen with 7 percent air voids had been used for the test.

This implies that the dynamic modulus values backcalculated from the IDT resilient modulus test

are also in the acceptable variability range.

[INSERT TABLE 1 HERE]

5. EXPERIMENTAL STUDY USING FIELD-CORED MIXTURES

The dynamic modulus test using the uniaxial testing mode cannot be performed on field-cored

mixtures because of the problem associated with the specimen size requirements of the uniaxial

test. Its application is therefore mostly limited to fresh mixtures fabricated in the laboratory. As

an alternative to the uniaxial tension test, the methods to integrate the dynamic modulus test to

the IDT testing conditions, which has been used in testing both laboratory-made and field-cored

mixtures, have been continuously developed by many researchers (Zhang et al. 1997a and 1997b,

Kim et al. 2004, Kim et al. 2005, Lee and Kim 2009, and Kim and West 2010). To evaluate the

17
backcalculation methodology from laboratory tests using field cores, IDT resilient modulus and

dynamic modulus tests were performed. The procedure for the IDT resilient modulus test was the

same as previously described for the laboratory-made specimens. For the dynamic modulus test,

the method used by Lee and Kim (2009) and Kim and West (2010), which adopted the dynamic

testing procedure used in AASHTO TP 62, was applied for measuring the dynamic modulus

values of IDT specimens.

5.1. Material Preparation and Testing on Field-Cored Mixtures

A total of twenty-seven field cores obtained from three field sections with different mixtures

were carefully cut to a diameter of 150 mm and a thickness of 38 mm. Each of the three mixtures

consisted of nine specimens. Three specimens were considered as one set, and each set was

assigned to each of the following three temperatures: 0°C, 10°C, and 20°C. For each temperature,

the three specimens were placed in the target temperature for at least 12 hours prior to the IDT

tests. After the specimens had been equilibrated with the specified testing temperature, the

resilient modulus test was conducted for five cycles in which each cycle consisted of a 0.1-

second haversine loading followed by a 0.9-second rest period. Dynamic modulus tests were

then conducted at five frequencies—0.1 Hz, 0.5 Hz, 1.0 Hz, 5.0 Hz, and 10.0 Hz—targeting a

horizontal strain level between 40 µe and 60 µe.

5.2. Dynamic Modulus from the IDT Dynamic Modulus Test

From the plane-stress constitutive equation for a linear elastic material shown in Equation (34),

the constitutive equation for a viscoelastic material in the frequency domain can be obtained by

18
replacing D and m in Equation (34) by their viscoelastic counterparts (Wineman and Rajagopal,

2000):

[
e x* = D * × s x0 - m *s y0 ] (44-1)

e *y = D * × [s y0 - m *s x0 ] (44-2)

where e x* and e *y are the complex amplitudes of the horizontal and vertical strains, s x0 and s y0

are the horizontal and vertical stress amplitudes, and D * and m * are the viscoelastic, complex

compliance and Poisson’s ratio. Combining Equations (44-1), (44-2), and (32) by eliminating

m * yields a single equation for D * :

*
e x* + be *y
D (w ) = (45)
s x0 + bs y0

Based on the theory of viscoelasticity, the horizontal and vertical complex strain amplitudes can

also be expressed in a form of Euler’s formula or complex numbers such that:

e x* (w ) = e x0 × e - id = e x0 × (cos(d x ) - i sin (d x ))
x
(46-1)

= e y0 × (cos(d y ) - i sin (d y ))
- id y
e *y (w ) = e y0 × e (46-2)

where e x0 and e y0 are the amplitudes of the horizontal and vertical strains that lag behind their

respective stresses with phase angles d x and d y . In a previous study conducted by the authors, it

was shown that the imaginary part of the viscoelastic Poisson’s ratio was negligible, indicating

that the horizontal and vertical phase angles were essentially identical to each other such that

(Lee and Kim 2009):

d =dx =dy (47)

19
Although it is not shown in this paper for the sake of brevity, it has also been observed that the

horizontal phase angle was almost identical to the vertical phase angle (i.e., the imaginary part of

the complex Poisson’s ratio is zero) for all mixtures tested in this study. Substituting Equations

(46) and (47) into Equation (45) yields:

e y0 + be y0 e y0 + be y0
D * (w ) = D'+iD' ' = × e -id = × (cos d - i sin d ) = D * × (cos d - i sin d ) (48)
s y0 + bs x0 s y0 + bs x0

As mentioned, taking the reciprocal of the dynamic compliance, |D*|, obtained from the above

equation, results in the dynamic modulus, |E*|, in the IDT testing mode. For simplicity, the

dynamic modulus obtained from the dynamic modulus test and from Equation (48) will be

referred to as the “measured” dynamic modulus in the subsequent section of the paper.

5.3. Comparison of Backcalculated Dynamic Modulus and Measured Dynamic Modulus in


the IDT test

Figure 7 shows the dynamic modulus |E*| values measured for all three mixtures compared to

those backcalculated using the data from the IDT resilient modulus tests. A correlation evaluated

by means of R2 for all mixtures at all levels of temperature and frequency was 0.993. The

average error of the backcalculated dynamic modulus was slightly less than 5 percent, and the

slope of the linear regression line was 0.984, which was very close to the equality line. This

proves that the backcalculation technique proposed in this paper is also effective for field-cored

mixtures.

[INSERT FIGURE 7 HERE]

To identify the effects of different mixtures, temperatures, and frequencies on the

backcalculated results, three scatter plots are provided in Figure 7, with each plot showing the

breakdown of the above variables in different series. Also shown in the figures are the slope

20
and the R2 values of each set of data at the given condition. For all cases, no significant change

in the slope of the regression line or in the R2 was observed. These observations lead to the

conclusion that the dynamic modulus can be backcalculated from the resilient modulus test data.

The approach developed in this study would significantly save time and effort in re-evaluating

the dynamic modulus of asphalt mixtures for which the resilient modulus test data are available.

6. SUMMARY AND CONCLUSIONS

In this study, a methodology was developed that allows for backcalculating the fundamental

viscoelastic material property, that is, creep compliance from the random loading and its

corresponding response. The creep compliance in terms of the power function and Prony series

could be backcalculated from the commonly used uniaxial and IDT resilient modulus test data

using the algorithms derived from the viscoelastic constitutive relations. The verification study

performed on the uniaxial resilient modulus test data clearly showed that the predictive strain

responses agreed well with known strain history. This indicates that the proposed methodology

was successful in backcalculating the creep compliance of asphalt mixtures subjected to any

random loading and could be extended to other testing modes, such as the IDT test.

For the mixture fabricated in the laboratory, the dynamic modulus and resilient modulus

tests in the uniaxial testing mode were performed in the testing setup specified in AASHTO TP

62. After the uniaxial test, the IDT resilient modulus test was conducted for the same mixture.

Backcalculated creep compliance was converted to complex compliance in the frequency domain

and then converted to the dynamic modulus. The backcalculated moduli from the uniaxial

resilient modulus test were slightly higher than the measured values, but only a 7-percent

21
difference in average dynamic modulus was observed. The dynamic modulus values

backcalculated from the IDT resilient modulus test were 23 percent lower than those measured in

the uniaxial test because the air void content of the IDT specimens was 3 percent lower than that

of the mixtures used in the uniaxial test. The reduction of 15 percent in dynamic modulus could

be estimated through the well-known predictive equation, indicating that the dynamic moduli

backcalculated from the IDT resilient modulus test were also in the generally accepted variability

range of 10 percent for asphalt mixture tests.

Because the dynamic modulus test using the uniaxial testing mode cannot be performed

on field-cored mixtures, the IDT dynamic modulus tests were performed for field-cored

specimens to evaluate the predictive capability of the backcalculation approach. Three mixtures

consisting of nine specimens were tested in the IDT testing mode at three temperatures—0°C,

10°C, and 20°C. The correlation evaluated by means of R2 for all mixtures at all levels of

temperature and frequency used was 0.993. The average difference of the backcalculated

dynamic modulus over the measured dynamic modulus was slightly less than 5 percent, and the

slope was 0.984, which was very close to the equality line. This indicates that the

backcalculation technique is also an effective approach for field-cored mixtures being evaluated

in the IDT testing mode.

Although the Mechanistic-Empirical Pavement Design Guide (MEPDG) could not

endorse the use of the resilient test protocol as the primary means of asphalt concrete modulus

characterization in the design of flexible pavements, it has been a primary mixture test, and a

considerable amount of laboratory testing has been done to date.

22
Because of the practical nature of the proposed backcalculation methodology, the

approach developed in this study would significantly save time and effort in re-evaluating the

dynamic modulus of asphalt mixtures for which the resilient modulus test data are available.

23
REFERENCES
American Association of State Highway and Transportation Officials, 1993. AASHTO Guide for
Design of Pavement Structures. American Association of State Highway and
Transportation Officials, Washington, D.C., U.S.A.
ARA., ERES Division, 2004. Guide for Mechanistic-Empirical Design of New and Rehabilitated
Pavement Structures. NCHRP 1-37A, Transportation Research Board, National Research
Council.
Buttlar, W. G., and Roque, R., 1994. Development and evaluation of the strategic highway
research program measurement and analysis system for indirect tensile testing at low
temperatures. Transportation Research Record, 1454, Transportation Research Board,
163-171.
Christensen, D. W., and Anderson D. A., 1992. Interpretation of Dynamic Mechanical Test Data
for Paving Grade Asphalt Cements. Journal of Association of Asphalt Paving
Technologists, 61, 67-116.
Findley, W. N., Lai, J. S., and Onaran, K., 1976. Creep and relaxation of nonlinear viscoelastic
materials, Dover Publications, Inc. NY.
Hondros, G., 1959. The evaluation of Poisson’s ratio and the modulus of materials of a low
tensile resistance by the brazilian (indirect tensile) test with particular reference to
concrete. Australian Journal of Applied Science, 10(3), 243-268.
Huang, Y. H., 1993. Pavement Analysis and Design, Prentice-Hall, NJ.
Kim, Y. R, Seo, Y., King, M., and Momen, M., 2004. Development and evaluation of the
strategic highway research program measurement and analysis system for indirect tensile
testing at low temperatures. Transportation Research Record, 1891, Transportation
Research Board, 163-173.
Kim, J., Roque, R., and Birgisson, B., 2005. Obtaining creep compliance parameters accurately
from static or cyclic creep tests. ASTM STP 1469, ASTM International, West
Conshohocken, PA, 177-197.
Kim, J., Sholar, G., and Kim, S., 2008. Determination of accurate creep compliance and
relaxation modulus at a single temperature for viscoelastic solids. Journal of Materials in
Civil Engineering, ASCE, 20(2), 147-156.
Kim, J. and West, R. C., 2010. Application of the viscoelastic continuum damage model to the
indirect tension test at a single temperature. Journal of Engineering Mechanics ASCE,
136(4).
Lee, H. and Kim, J., 2009. Determination of viscoelastic Poisson’s ratio and creep compliance
from the indirect tension test.” Journal of Materials in Civil Engineering ASCE, 21(8),
416-425.
Park, S. W., and Schapery, R. A., 1999. Methods of interconversion between linear viscoelastic
material functions. Part I – a numerical method based on Prony Series. International
Journal of Solids and Structures, 36, 1653-1675.
Park, S. W., and Kim, Y. R., 2001. ‘Fitting prony-series viscoelastic models with power-law
presmoothing. Journal of Materials in Civil Engineering, ASCE, 13(1), 26–32.
Roque, R., and Buttlar, W. G., 1992. The development of a measurement and analysis system to
accurately determine asphalt concrete properties using the indirect tensile mode. Journal
of the Association of Asphalt Paving Technologists, 61, 304-332.

24
Roque, R., Buttlar, W. G., Ruth, B. E., Tia, M., Dickison, S. W., and Reid, B., 1997. Evaluation
of SHRP indirect tension tester to mitigate cracking in asphalt concrete pavements and
overlays. Final Report, FDOT B-9885, University of Florida, Gainesville, FL.
Schapery, R. A., and Park, S.W., 1999. Methods of interconversion between linear viscoelastic
material functions. Part II – an approximate analytical method. International Journal of
Solids and Structures, 36, 1677-1699.
Tschoegl, N. W., 1989. The phenomenological theory of linear viscoelastic behavior: an
introduction. Springer-Verlag, NY.
Tschoegl, N. W., Knauss, W. G., and Emri, I., 2002. Poisson’s ratio in viscoelasticity: a critical
review. Mechanics of Time-Dependent Materials, 6, 3-51.
Wineman, A. S., and Rajagopal, K. R., 2000. Mechanical response of polymers: an introduction.
Cambridge University Press, NY.
Witczak, M. W., Kaloush, K., Pellinen, T., El-Basyouny, M., and Von Quintus, H., 2002. Simple
performance test for superpave mix design, Report No. 465, Transportation Research
Board, National Research Council, Washington, DC.
Zhang, W., Drescher, A., and Newcomb, D. E., 1997a. Viscoelastic analysis of diametral
compression of asphalt concrete. Journal of Engineering Mechanics, ASCE, 123(6), 596-
603.
Zhang, W., Drescher, A., and Newcomb, D. E., 1997b. Viscoelastic behavior of asphalt
concrete in diametral compression. Journal of Transportation Engineering, ASCE,
123(6), 495-502.

25
List of Tables and Figures

Table 1. Volumetric properties of the laboratory-fabricated mixture

Figure 1. Schematic illustration of the uniaxial test

Figure 2. (a) Schematic illustration of the IDT test and (b) plane stress distributions

Figure 3. (a) Uniaxial test setup and (b) specimen dimensions

Figure 4. (a) Indirect tensile test setup and (b) specimen dimensions

Figure 5. Measured vs. predicted uniaxial strains for (a) specimen 1 and (b) specimen 2

Figure 6. Measured vs. backcalculated dynamic moduli

Figure 7. Scatter plots of measured vs. backcalculated IDT dynamic moduli with different series
representing different (a) mixtures, (b) temperatures, and (c) frequencies

26

You might also like