Download as pdf or txt
Download as pdf or txt
You are on page 1of 38

11451: Pavement Mechanics

Handout Notes

Asphalt Pavements

Eyal Levenberg

1. Layered Elastic Half-space .......................................................................................... 1


1.1 Axisymmetry ........................................................................................................ 1
1.2 Hankel transform .................................................................................................. 3
1.3 Equations .............................................................................................................. 4
1.4 Numerical considerations ..................................................................................... 6
1.5 Multiple and moving loads ................................................................................... 7
2. Elastostatic FWD Backcalculation .............................................................................. 9

Linear Viscoelastic Solids


3. Boltzmann Superposition ........................................................................................... 11
3.1 One-dimensional creep formulation ................................................................... 13
3.2 One-dimensional relaxation formulation ............................................................ 16
3.3 Three-dimensional formulation .......................................................................... 17
4. Interconversion and Laplace Transform .................................................................... 19
5. Frequency Domain ..................................................................................................... 23
5.1 Storage and loss moduli ...................................................................................... 24
5.2 The complex modulus ........................................................................................ 26
6. Time-Temperature Superposition .............................................................................. 29
7. Mechanical Analogs and Spectra ............................................................................... 31

Appendix: Bibliography ..................................................................................................... 37

DTU Civil Engineering – Section for Geotechnics and Geology – August 2021
Pavement Mechanics (11451) August 2021

1. Layered Elastic Half-space

The standard layered elastic formulation involves a semi-infinite medium composed of I  1


parallel (horizontal) layers attached to a half-space (see Figure 1). The layers are numbered

serially, with the top layer being Layer 1 and the half-space, Layer I . Each layer is

characterized by a Young's modulus Ei , Poisson's ratio  i , and thickness hi , with the subscript

i serving as a layer identifier. The depth from the surface to the interface separating layers i

and i  1 is denoted by zi . Accordingly, z1 is the thickness of Layer 1 (i.e., z1  h1 ), z 2 is the

combined thickness of Layers 1 and 2 (i.e., z2  h1  h2 ), and so on. The collective thickness of

all I  1 layers is denoted by H , i.e., H  zI 1  h1  h2  ...  hI 1 . Vertical loading,


representing tire-pavement contact interaction, operates at the surface (i.e., top of Layer 1). It

entails a uniform stress with intensity q0 applied over a circular area with radius a .

Figure 1. Cross-sectional view of a standard layered elastic pavement model.

1.1 Axisymmetry

We first note that for a single circular load the problem is axisymmetric. Conveniently, a

cylindrical coordinate system is introduced such that its origin is positioned at the surface of

Layer 1 just beneath the loaded area; the vertical z -axis is drawn into the medium and the

radial r -axis is parallel to the top boundary. The medium displacements in the z , r and 

directions for layer i are respectively denoted by: (u z ) i , (u r ) i , and (u )i .

Because the problem is axisymmetric, the following simplifications apply

© 1999–2021 Eyal Levenberg 1


Pavement Mechanics (11451) August 2021

 (u r ) i  (u z ) i
(u )i  0 0 0 (1)
 

Also, the kinematic equations, relating strains to displacements, reduce to the following set
 (u z )i
( z )i  (2)
z
 (ur )i
( r )i  (3)
r
(ur )i
( )i  (4)
r
1   (u z )i  (ur )i 
( zr )i  ( rz )i     (5)
2  r z 
( r )i  ( r )i  ( z )i  ( z )i  0 (6)

where ( z ) i , ( r ) i , (  ) i , ( rz )i , ( r )i , and ( z )i are the strains in z , r ,  , r  z , z   , and


z   directions for layer i (respectively). The corresponding constitutive (stress-strain)

equations are
1  
( r )i  ( r )i  i (  )i  i ( z )i (7)
Ei Ei Ei
1  
(  )i  (  )i  i ( r )i  i ( z )i (8)
Ei Ei Ei
1  
( z )i  ( z )i  i (  )i  i ( r )i (9)
Ei Ei Ei
1  i
( rz )i  ( rz )i (10)
Ei

where ( r )i , (  )i , ( z )i , and ( rz )i are the respective stresses in the z , r ,  , and r  z

directions for layer i .

Moreover, we are assuming a weightless medium without body forces; hence, the

equilibrium equations are reduced to


 ( r )i  ( zr )i ( r )i  (  )i
  0 (11)
r z r
 ( zr )i  ( z )i ( zr )i
  0 (12)
r z r

Recalling that ( r )i  0 and also that ( z )i  0 .

© 1999–2021 Eyal Levenberg 2


Pavement Mechanics (11451) August 2021

1.2 Hankel transform

The zero-order Hankel transform maps a function f (r ) into a function F (m) according to

F (m)   rJ (mr ) f (r )dr
r 0
0 (13)

in which J 0 () is the Bessel function of the first kind of order zero. The Hankel transform is

self-reciprocal, with the inverse transform of F (m) to yield f (r ) calculated as follows



f (r )  
m 0
mJ 0 (mr ) F (m)dm (14)

Say f (r ) represents a distributed circular load with intensity q0 and radius a


f (r)  q0 H (a  r ) (r  0) (15)

in which H ( x) is the Heaviside step function defined as


H ( x)  0 for x  0
H ( x)  0.5 for x  0 (16)
H ( x)  1 for x  0

Graphically, the Heaviside step function is represented as follows (with   0 ):

Figure 2. Graphical representation of the Heaviside step function (   0 ).

Substitution of f (r ) in Equation 15 into Equation 13 gives


 a
q0 aJ1 (ma)
F (m)  
r 0
rJ 0 (mr )q0 H (a  r )dr  q0  rJ (mr )dr 
r 0
0
m
(17)

where J 1 () is the Bessel function of the first kind of order one, and the following relations

were employed

rJ1 (mr )
 rJ 0 (mr )dr 
m
and J1 (0)  0 (18)

Now, using the inverse Hankel transform in Equation 14, we get

© 1999–2021 Eyal Levenberg 3


Pavement Mechanics (11451) August 2021


f (r )  q0 H (a  r )  q0 a 
m 0
J 0 (mr ) J1 (ma)dm (r  0) (19)

Class Exercise #3: Numerical Hankel Transform

1.3 Equations

According to Augustus Edward Hough Love (1927), a scalar potential function  (r , z ) (units

of force×length) that satisfies the biharmonic equation 22  0 , where  2 is the

axisymmetric Laplace operator ( 2   2 / r 2  r 1 / r   2 / z 2 ) may be used to derive

stresses and displacements in an elastic medium under axisymmetric conditions (compatibility

is guaranteed). The approach is applied to layer i , i.e., a different scalar potential function  i
is assigned to each layer

  2i 
( z )i   (2  ) 2
   (20)
z  z 2 
i i

  2i 
( r )i  
 i  2
   (21)
z  r 2 
i

  1 i 
(  )i   i  i 
2
 (22)
z  r r 

  2i 
( rz )i    i   
2
1  (23)
r  z 2 
i

1  i   2i 1 i 
(uz )i      i    
2
1 2  (24)
r 2 r r 
i
Ei 

1  i   2i 
(ur )i    (25)
Ei  rz 

The above set embodies soil mechanics sign convention, that positive stress or strain indicate

compressive conditions.

For the case of layered elastic half-space, a suitable expression for such a potential

function was first proposed by Burmister (1943), see also Huang (2004) and Van Cauwelaert

(2003)

© 1999–2021 Eyal Levenberg 4


Pavement Mechanics (11451) August 2021

 m z  z   m z  zi1 
J 0 (mr ) J1 (ma )  Ai e i  Bi e 

i  q0 aH  3
  dm (26)
m3  C mze m zi  z   D mze  m z  zi1  
m 0  i i 

in which a  a / H , r  r / H , z  z / H , and zi  zi / H are dimensionless entities (with


z0  0 and z I   ), while Ai  Ai (m) , Bi  Bi (m) , Ci  Ci (m) , and Di  Di (m) are unitless

functions of m that embody the layered system properties and connectivity.

Substitution of  i into the stress and displacement expressions (Equations 20-25) yields

the following
    Ai  Ci 1  2 i  mz   e m zi  z  
( z )i  q0 a  J 0 (mr )   J1 (ma )dm (27)
   Bi  Di 1  2 i  mz   e m z  zi1  
m 0
 
   Ai  Ci 1  mz   e  m zi  z   

  J ( mr ) 
J ( mr )  

1
 0
  i 
( r )i  q0 a    mr    B  D 1  mz  e  m z  zi 1   

 i   J1 (ma )dm (28)

 
m 0
 2 J (mr ) C e  m zi  z   D e m z  zi1  
 i 0 i i


J ( mr )   Ai  Ci 1  mz   e m zi  z   
   
1

  i 
  m z  zi 1   
i
(  )i  q0 a  mr   B  D 1  mz e
   J1 (ma )dm (29)

 
m 0
 2 J (mr ) C e  m zi  z   D e  m z  zi1  
 i 0 i i

   Ai  Ci  2 i  mz   e  m zi  z  
( rz )i  q0 a  J1 (mr )   J1 (ma )dm (30)
   Bi  Di  2 i  mz   e m z  zi1  
m 0
 

q0 aH 1  i   J 0 (mr )   Ai  Ci  2  4 i  mz   e
 m zi  z 

  J1 (ma )dm
(u z )i 
Ei  m    B  D  2  4  mz   e m z  zi1   (31)
m 0
 i i i 

q0 aH 1  i   J1 (mr )   Ai  Ci 1  mz   e
 m zi  z 

  J (ma )dm
(ur )i    (32)
   Bi  Di 1  mz   e
Ei m   m z  zi 1   1
m 0

wherein the m functions Ai , Bi , Ci , and Di for i  1, 2,..., I are to be obtained by imposing

boundary conditions. Initially, for the surface of the top layer, where i  1 and z  0 , vertical

stress must equal the applied stress distribution, and shear stresses must vanish everywhere
( z )1  q (r ) (33)
( rz )1  0 (34)

wherein, based on Equation 19, q(r ) is simply given by

© 1999–2021 Eyal Levenberg 5


Pavement Mechanics (11451) August 2021


q(r )  q0 a 
m 0
J 0 (mr ) J1 (ma )dm (35)

Second, at the bottom of a layer i and at the top of a layer i  1 , i.e., at the interface where
z  zi , vertical stresses, vertical displacements, and shear stresses must be identical
( z )i  ( z )i 1 (36)
(uz )i  (uz )i 1 (37)
( rz )i  ( rz )i 1 (38)

As for radial displacements at the interface where z  zi , perfect bonding essentially means

(ur )i  (ur )i 1 . Alternatively, partial (or imperfect) bonding can be introduced via a bonding
parameter  i (positive, units of force/length³) such that
( rz )i   i  (ur )i  (ur )i 1  (39)

or equivalently
( rz )i
(ur )i  (ur )i 1  (40)
i

In these expressions, as i   full bond conditions are modeled because radial displacements

above and below the interface are forced to be identical. Also, as i  0 ideal slippery

interface conditions are modeled, because the shear stress across the interface is forced to zero.

Third, the sought solution must embody the requirement that all responses vanish for i  I , as

the evaluation depth approaches infinity; i.e., z   . Formally, this condition means
AI  CI  0 (41)

Thus, the functions Ai , Bi , Ci and Di are determined from solving Equations 33, 34, 36-38,

39 or 40, and 41 for m values ranging from zero to infinity. Then after, stresses and

displacements are determined from Equations 27-32; finally, strains are calculated from

Equations 7-10.

1.4 Numerical considerations

The integration over m in Equations 27-32 is to be performed numerically. In general, selecting


a small m increases the accuracy of the result but also increases the computational cost

© 1999–2021 Eyal Levenberg 6


Pavement Mechanics (11451) August 2021

involved. This is because each integration step (each new m ) requires inverting a matrix. In
addition, the oscillations of the Bessel functions require special attention; it is therefore

recommended to perform the integration using Gauss formulas – applied between the zeros of

the participating Bessel functions. Also, there is a need to address the problem of slow

convergence when evaluating responses at points close to the surface. For this purpose,

acceleration techniques such as the double exponential formula or Richardson extrapolation

(or both) may be used.

Project #1: Layered Elastic Half-Space

1.5 Multiple and moving loads

To address multiple loads, we need to transform the responses from a cylindrical coordinate

system into a (right-handed) Cartesian coordinate system. The angle  is measured from the

positive direction of the x -axis to that of the y -axis.

x
 
r
r
r
y


(Top view)

 x  xy  xz   r  r  rz 
  T  
 xy  y  yz    S     r     z    S 
 xz  yz  z   zr  z  z 
 

u x   ur   cos sin  0
  T  
u y    S   u   S    sin  cos 0 
u  u   0 1 
 z  z 0

Figure 3. Coordinate transformation from Cylindrical to Cartesian.


© 1999–2021 Eyal Levenberg 7
Pavement Mechanics (11451) August 2021

The effects of moving loads may be analyzed using the quasi-static approximation. Under

this approach, the loading is assumed to be slow enough so that inertia effects can be

disregarded and a moving load is 'animated' using the static solution. Beyond a certain speed

limit, the effects of the stress waves traveling through the system cannot be overlooked and the

quasi-static approximation loses its accuracy and relevancy.

Class Exercise #4: Layered Elasticity

© 1999–2021 Eyal Levenberg 8


Pavement Mechanics (11451) August 2021

2. Elastostatic FWD Backcalculation

The Falling Weight Deflectometer (FWD) is a nondestructive pavement testing device; it

applies an impulse load to the pavement surface through a circular plate. The load is generated

by dropping a mass onto the plate, and the resulting deflections are measured by a series of

geophones positioned at different distances from the load plate center.

Figure 4. Typical FWD load plate and geophone arrangement.

In a usual FWD measurement campaign, only peak load level and peak deflection values

are collected and reported for a given test location. Nonetheless, the device can capture the

entire time history of the loading and deflections for a duration of about 120 milliseconds (see

Figure 5)

Figure 5. FWD load-deflection history.

© 1999–2021 Eyal Levenberg 9


Pavement Mechanics (11451) August 2021

To understand what can be done with FWD readings, we can look at what influences the

measured deflections. Included are the following: (i) loading conditions: force level, drop

height, load plate size, stress distribution; (ii) pavement composition (structure and subgrade):

geometry, layer thicknesses, cracks and discontinuities; (iii) mechanical properties of the

layers; and (iv) interface attributes.

Backcalculation refers to the process of determining the mechanical properties of layers

in a given 'model-pavement' using the measured surface deflections. Currently, the most

common approach for backcalculation is performed using layered elastic (isotropic) theory.

This is done "backwards" with a set of seed moduli (assuming the Poisson's ratios) and

"known" layer thicknesses. For the calculations, only the response peaks are typically used,

i.e., the peak FWD load along with the deflection peaks. An optimization algorithm is applied

to manipulate the moduli values until an error function representing the difference between

measured and computed deflections is minimized. This approach is called 'elastostatic

backcalculation' because we disregard wave propagation effects (time dependence in material

properties is also disregarded). Common computer programs that perform elastostatic

backcalculation include: (i) Elmod (Dynatest); (ii) ERIDA (KUAB); (iii) MODCOMP (Cornell

University); (iv) MODULUS (Texas DOT); (v) EVERCALC (Washington DOT); (vi)

CalBack (California).

There are more advanced backcalculation methods, employing more advanced structural
and constitutive models, attempting to compute more realistic/representative layer properties.

These methods try to make use of the entire measured time-history. Perhaps due to their high

level of sophistication, they are not commonly used in engineering practice. Often, FWD test

data are used without modeling. Such direct usage includes: (i) observing deflection variability

(spatial, seasonal); (ii) joint load transfer efficiency in concrete pavements; and (iii) void

detection.
Class Exercise #5: Backcalculation

Project #2: Backcalculation

© 1999–2021 Eyal Levenberg 10


Pavement Mechanics (11451) August 2021

Linear Viscoelastic Solids

3. Boltzmann Superposition

Linear (isotropic) elasticity is considered the simplest type of constitutive model for pavement

layers. Mathematically, it implies that the state of strain at any given time in the material (or

structure) is directly proportional to the stress state at that time; the constants of proportionality

being the elastic properties. In reality, however, pavement materials (and especially asphalt

concrete), display time-dependent mechanical behavior; to list a few examples: (i) the material

creeps under a constant level of load, (ii) the material relaxes under a constant level of

deformation, and (iii) the material exhibits dependence of strain on stress-rate. A more realistic

(improved) constitutive theory, that can mathematically capture and reproduce these properties,

is the theory of linear viscoelasticity first formulated by Boltzmann in 1874. According to the

theory, any 'response' of interest R in a viscoelastic body (e.g., stress state or displacement at

a point) is a function of the entire 'input' history I (e.g., boundary force or displacement) up to

the present time, and not simply a function of the current value of I ; mathematically, this can

be written as
R  R I  (42)

in which the curly brackets are used to denote the fact that the present value of R is a function

of the entire history of I ( R is said to be a functional of I ). A linear dependence of R on I

means that the superposition criterion is satisfied


R  I a  I b   R I a   R I b  (43)
where I a and I b are either the same or different input histories.

Boltzmann offered a mathematical operator that can reproduce linearity, and compute a

'response' considering the entire 'input' history up to the current time


 t
R(t )  
 
RH (t   )dI ( ) (44)

in which t is current/calculation time,  is a time-like integration variable, R(t ) is the


computed 'response' at time t , and I ( ) is the 'input' applied at time  . The lower integration

© 1999–2021 Eyal Levenberg 11


Pavement Mechanics (11451) August 2021

limit can be changed to   0 instead of    whenever the input I ( ) equals zero or does
not exist for negative's. The time function RH (t ) serves the task of a weighing function; it is

the kernel of the operator; the argument of RH is t  in order to generate a 'response' that

depends upon the time elapsed since the application of the 'input'. This operator is identified in

the technical literature by several names: (i) Boltzmann superposition; (ii) hereditary integral;

(iii) convolution integral; and (iv) Duhamel's integral.

The integral in Equation 44 is a Stieltjes type of integral, with


dI ( )  lim  I (   )  I ( )  (45)
 0

If the first time derivative of I ( ) is continuous, it can be re-written in the usual, Riemann

type, format
 t
dI ( )
R(t ) 



RH (t   )
d
d (46)

When the Heaviside unit step function is used as an input, i.e., I  H ( ) , we have
dI ( ) dH ( )
   ( ) (47)
d d

in which  ( ) is the Dirac delta function (or the impulse function), having the following

properties (with   0 )

 ( x)  0 for x  0 and   ( x)dx  1

(48)

 ( x)

1/ 


Figure 6. The Dirac delta function (   0 ).

Substitution of I  H ( ) into Equation 46 gives

© 1999–2021 Eyal Levenberg 12


Pavement Mechanics (11451) August 2021

 t
R(t )  
 
RH (t   ) ( )d  RH (t ) (49)

The result above means that the "weighing" function RH (t ) is equal to the material (or system)

response R when the input I is a unit step.

Class Exercise #6: Unit Step Input

3.1 One-dimensional creep formulation

Consider a one-dimensional loading of a linear viscoelastic material. The 'input' I is a stress

history  ( ) and the 'response' R at time t is the resulting strain  (t ) . In this case, the

Boltzmann superposition integral gives


 t
 (t )  
 
D(t   )d ( ) (50)

in which RH (t ) was replaced by D(t ) , the so-called 'creep compliance'; a material property to

be determined experimentally. Equation 50 is the one-dimensional viscoelastic creep

formulation. Based on the previous analysis (i.e., Equation 49), D(t ) may be directly
determined by applying a step stress input. This is the case for a creep test, where the applied

stress history is  (t )   0 H (t )

 ( )

0 

Figure 7. Applied stress history in a creep test.

Going back to the superposition integral, we have


 t  t
d ( )
 (t )   D(t   ) d   D(t   ) 0 ( )d  0 D(t ) (51)
  d  

It can be seen that D(t ) is directly obtained from the recorded strain  (t ) in the creep test

© 1999–2021 Eyal Levenberg 13


Pavement Mechanics (11451) August 2021

 (t )  unitless 
D(t )   stress units   stress units 
-1
(52)
0  

A typical strain outcome in a creep test is shown in Figure 8 wherein  0 is the instantaneous

strain response, and   is the equilibrium strain level.

 (t )



0 t

Figure 8. Typical strain-history response in a creep test.

Accordingly, the limiting values of D(t ) are


0 
lim D(t )  D0  and lim D(t )  D  (53)
t 0 0 t  0

If D(t ) is known, the creep formulation allows for computing the strain response to an

arbitrary-shaped stress history - not just a simple step. To better understand how this is

achieved, consider the case shown in Figure 9, dealing with a two-step stress history. The creep

stress 1 is applied at time 1 , and  2 is applied at time  2 , with  2  1 . The resulting

strain is also (schematically) shown

Figure 9. Two-step creep and strain response.

© 1999–2021 Eyal Levenberg 14


Pavement Mechanics (11451) August 2021

For the time interval  1  t   2 we have


 (t )  D(t 1 )1 (54)

and for the time interval t   2 , we have


 (t )  D(t 1 )1  D(t  2 ) 2 (55)

Alternatively, we can write  as a function of time  , i.e.,    ( )


1   (1 ) and  2   ( 2 ) (56)

which gives
 (t )  D(t 1 ) (1 )  D(t  2 ) ( 2 ) (57)

and finally, for i creep 'steps'


 (t )   D(t   i ) ( i ) (58)
i

The above summation can be viewed as the discrete (series) version of the original integral
 t
 (t ) 



D(t   )d ( ) (59)

Class Exercise #7: Viscoelastic Response

Here is a graphical (log-log scale) example of D(t ) vs. t for an asphalt concrete specimen at a

temperature of 30˚C

Figure 10. Example of an asphalt concrete creep compliance master curve.

© 1999–2021 Eyal Levenberg 15


Pavement Mechanics (11451) August 2021

The fact that the D(t ) curve levels off towards a finite value as t   means the material is a
viscoelastic solid. If D(t ) were to increase indefinitely with time, i.e., limt  D(t )   , then

we would have a viscoelastic fluid.

Assignment #2: One-Dimensional Linear Viscoelastic Strain Response

3.2 One-dimensional relaxation formulation

Let as consider (again) a one-dimensional loading of a linear viscoelastic solid. This time, the

'input' I is a strain history  ( ) and the "response" R at time t is the resulting stress  (t ) .

The Boltzmann superposition gives in this case


 t
 (t )  
 
E (t   )d  ( ) (60)

here, E (t ) is the relaxation modulus, a complimentary material property to the creep

compliance D(t ) . This is the one-dimensional viscoelastic relaxation formulation. As before,


E(t ) can be determined from a relaxation experiment in which a step-strain history is applied:
 (t )   0 H (t ) and the resulting stress recorded.

 ( )

0 

Figure 11. Applied strain history in a relaxation test.

Going back to the superposition integral, we have


 t  t
d  ( )
 (t )   E (t   ) d   E (t   ) 0 ( )d  0 E (t ) (61)
  d  

from which it can be seen that E (t ) may be directly obtained from the recorded stress in the

relaxation experiment

© 1999–2021 Eyal Levenberg 16


Pavement Mechanics (11451) August 2021

 (t )  stress units 
E (t ) 
0  unitless   stress units (62)

A typical stress response in a relaxation test is

Figure 12. Typical stress-history response in a relaxation test.

The limiting values of E (t ) are


lim E (t )  E0 and lim E (t )  E (63)
t 0 t 

Note that if the material properties were time-independent, the relaxation modulus would

become Young's modulus E (or loosely: elasticity modulus), in which case we would have
E  1/ D . However, for viscoelastic materials, this relationship does not hold, i.e.,

E(t )  1/ D(t ) .

3.3 Three-dimensional formulation

At this point, let us extend the one-dimensional treatment onto the multiaxial case. The "input"

is a stress history  kl ( ) and the "output" at time t is the resulting strain  ij (t ) . The Boltzmann

superposition gives
t
 ij (t ) 


Dijkl (t   )d kl ( ) (i, j, k , l  1..3) (64)

in which Dijkl (t ) is a tensor of creep functions and the summation convention is used in which

repeated indices are summed over their range.

The above expression can be written using single index notation


 t
 i (t ) 
 
 Dij (t   )d j ( ) (i, j  1..6) (65)

© 1999–2021 Eyal Levenberg 17


Pavement Mechanics (11451) August 2021

Similar to linear isotropic elasticity where there are two independent material constants

capturing the material behavior ( E and  ), in the case of linear isotropic viscoelasticity there
are two independent creep time functions:

 D11 (t ) D12 (t ) D12 (t ) 


 D (t ) D (t ) D (t ) 
 12 11 12 
 D12 (t ) D12 (t ) D11 (t ) 
Dij (t )   
 2( D11 (t )  D12 (t )) 
 2( D11 (t )  D12 (t )) 
 
 2( D11 (t )  D12 (t )) 

One practical way to simplify this representation is to write Dij (t ) as a product of two separate

terms
Dij (t )  Pijve  D(t ) (66)

in which Pijve is composed of unitless engineering parameters that are explicitly independent of

time and the scalar time-function D(t ) is the linear one-dimensional viscoelastic creep-

compliance. The consequence of this simplification is that one unique creep (time) function

controls the viscoelastic deformation in all directions; this has been experimentally validated

for asphalt concrete.

Similarly, it is possible to write a three-dimensional relaxation formulation


 t
 i (t ) 



Eij (t   )d  j ( ) (i, j  1..6) (67)

© 1999–2021 Eyal Levenberg 18


Pavement Mechanics (11451) August 2021

4. Interconversion and Laplace Transform

Referring to the creep formulation, suppose we would like to apply a stress input  (t ) such
that the strain response is a unit step H (t ) . This can be achieved if the applied stress history

resembles E (t ) - because E (t ) is by definition the stress response to a unit step strain


 ( )  H ( ) . Using this reasoning we can establish the classic relation between D(t ) and E(t )
dE ( )
t t



0
D(t   )dE ( ) 

 D(t  )
0
d
d  H (t ) (68)

In order to formally derive the relationship between the creep compliance and relaxation

modulus (along with other useful results), the Laplace transform is used. If f (t ) is a continuous

'time' function ( t  0 ) that is of exponential order1 as t   , its Laplace transform L  f (t ) is

f (s) , defined as

L  f (t )  f ( s)   f (t )e st dt (69)
t 0

As can be seen, a function of t is transformed into a (different) function of s . Some noteworthy

Laplace transform properties:

(i) The transform is unique;

(ii) Linearity
L c1 f1 (t )  c2 f 2 (t )  c1 f1 ( s)  c2 f 2 ( s) (70)

(iii) Translation theorem:

Translation of the "real" function by an amount a


L  f (t  a) H (t  a)  e  as f (s ) (71)

Translation of the "transformed" function by an amount a

L  f (t )e at   f (s  a) (72)

(iv) Transform of a derivative ( f discontinuous at t  0 )


 df (t )  
L   sf ( s )  f (0 ) (73)
 dt 

1
A function is of exponential order if e at f (t )  M with a and M positive constants.

© 1999–2021 Eyal Levenberg 19


Pavement Mechanics (11451) August 2021

(v) Transform of a definite integral

 t  f ( s )
L   f ( )d   (74)
 0  s

(vi) Transform of a convolution (Borel's theorem):

Convolution of the two functions f1 (t ) and f 2 (t ) is given by the expression


t



0
f1 (t   ) f 2 ( )d (75)

Laplace transform of the convolution integral is given by

 t 
L   f1 (t   ) f 2 ( )d   L  f1 (t ) L  f 2 (t )  f1 ( s ) f 2 ( s ) (76)
 0 
Hence, since f1 (s) f 2 (s) is commutative, it follows that the convolution expression is also

commutative
t t


 0
f1 (t   ) f 2 ( )d  
 0
f1 ( ) f 2 (t   )d (77)

(vii) Initial and final limit value theorems


lim f (t )  lim sf ( s) (78)
t 0 s 

lim f (t )  lim sf ( s) (79)


t  s 0

(viii) Useful transforms


f (t )   (t ) f (s)  1 (80)

f (t )   (t  a) f (s)  eas (81)

f (t )  H (t ) f (s)  1/ s (82)

e  as
f (t )  H (t  a) f ( s)  (83)
s

0
f (t )  et / 0 f ( s)  , 0  0
1 0s (84)

1
f (t )  1  et /0 f (s)  , 0  0
s (1   0 s ) (85)

f (t )  t f (s)  1/ s2 (86)

© 1999–2021 Eyal Levenberg 20


Pavement Mechanics (11451) August 2021

 ( n)
f (t )  t n1 f (s)  , n0 (87)
sn

1
f (t )  ei0t f ( s)  , 0  0
s  i0 (88)

0
f (t )  sin(0t ) f ( s)  , 0  0
s  02
2 (89)

s
f (t )  cos(0t ) f ( s)  , 0  0
s  02
2 (90)


* the Gamma function: (n)   x n1e x dx for integer n, (n)  (n  1)!
0

Class Exercise #8: Laplace Transform

Now, applying the Laplace transform to the creep formulation gives

 t d ( ) 
L  (t )   ( s)  L   D(t   ) d 
 0 d  (91)
 d (t ) 
 L D(t ) L    D( s) s ( s)
 dt 
or

 (s)  sD(s) (s) (92)

Similarly, applying the Laplace transform to the relaxation formulation gives

 d  ( ) 
t
L  (t )   ( s)  L   E (t   ) d 
 0 d  (93)
 d  (t ) 
 L E (t ) L    E ( s) s ( s)
 dt 
or
 ( s )  sE ( s) ( s) (94)

From the above two transformed equations, both  ( s ) and  ( s) can be eliminated to yield
1
D(s) E (s)  (95)
s2
Also, using the limit value theorems we find that

© 1999–2021 Eyal Levenberg 21


Pavement Mechanics (11451) August 2021

D0  lim D(t )  lim sD( s)


t 0 s 

 1  1
 lim   

s  sE ( s )
 lim sE ( s) (96)
s 

1 1
 
lim E (t ) E0
t 0

D  lim D(t )  lim sD( s)


t  s 0

 1  1
 lim  

s 0 sE ( s )
 lim sE ( s) (97)
s 0

1 1
 
lim E (t ) E
t 

Assignment #3: Analytic Interconversion

Project #3: Numerical Interconversion

© 1999–2021 Eyal Levenberg 22


Pavement Mechanics (11451) August 2021

5. Frequency Domain

When a linear viscoelastic solid is exposed to simple harmonic excitation (sine or cosine input

wave), the steady-state response attained after a 'long time' is also sinusoidal, having the same

frequency as the excitation but different amplitude and phase. To give a one-dimensional

example, let the 'input' be a strain history of the form  (t )   0 sin(t ) in which  0 is the
applied strain amplitude and  is the applied angular frequency (   2 f ). The resulting

steady-state stress response is  (t )   0 sin(t   ) in which  0 is the stress amplitude and 

is the phase angle by which the strain lags behind the stress (or the angle by which the stress

leads the strain).


t=/

Figure 13. Steady-state viscoelastic response to a sinusoidal excitation.

The stress amplitude  0 is a function of the applied frequency, i.e.,  0   0 ( ) . In general, as


 increases,  0 also increases. However, at both extremes, when   0 or    ,  0
approaches a finite value. Given that  0 is a frequency independent constant, it is instructive

to depict  0 /  0 vs.  . This "amplitude ratio" of stress to strain under steady-state sinusoidal

excitation is called 'dynamic modulus', usually denoted by E * . This term is a little misleading

since it has nothing to do with dynamic/inertia effects.


Log-scale

Log-scale

Figure 14. Norm of complex modulus vs. angular frequency (log-log scale).

© 1999–2021 Eyal Levenberg 23


Pavement Mechanics (11451) August 2021

The phase angle  also depends on the frequency level, i.e.,    ( ) with positive values up
to  / 2 (or 90 ). For elastic (time independent) solids the phase angle is identically zero, which

means that the stress and strain are always in-phase. However, for viscoelastic solids, the

dependency of  on  is such that elastic behavior is approached only at very high and very

low frequency levels:

Phase angle

Log-scale 

Figure 15. Phase angle vs. angular frequency (semi-log scale).

5.1 Storage and loss moduli

Using the trigonometric relation sin(   )  sin( ) cos( )  cos( )sin( ) it is possible to

decompose the steady-state stress response into two elements; the first in-phase with the

applied strain and the second 90 out-of-phase

 (t )   0 cos( )  sin(t )   0 sin( )  cos(t )


(98)
in-phase with 90 out-of-phase
appliedstrain with appliedstrain

Using the above expression, an in-phase modulus can be defined by taking the amplitude ratio

of the in-phase stress to strain. Similarly, an out-of-phase modulus can be defined by taking

the amplitude ratio of the out-of-phase stress to strain


0 0 E2
E1  cos( ) E2  sin( ) tan( )  (99)
0 0 E1

© 1999–2021 Eyal Levenberg 24


Pavement Mechanics (11451) August 2021

E1 is called "storage modulus" and E2 is called "loss modulus". Given that both  0 and  are

functions of  , we have E1  E1 () and E2  E2 ( ) . And finally, the steady-state stress

response can be written as


 (t )   0 E1 sin(t )   0 E2 cos(t ) (100)

Formally, the entire aforementioned derivation can be obtained by using the "relaxation"

formulation
 t
 (t ) 



E (t   )d  ( ) (101)

For later use, we decompose the relaxation modulus as follows E(t )  E  E(t ) such that
lim E (t )  0 (*Reminder: lim E (t )  E ) (102)
t  t 

Hence, the above Boltzmann integral is re-written


 t
 (t )  E (t )  
 
E (t   )d  ( ) (103)

Application of the harmonic excitation  ( )   0 sin( ) , from which we have


d ( )   0 cos( )d , gives upon substitution
 t
 (t )  E 0 sin(t )   0

 
E (t   ) cos( )d (104)

Changing the integration variable from  to s using t   s , from which we have ds  d

, generates a corresponding change of integration limits: s   (  ) and s  0(  t )

, and yields
s 
 (t )  E 0 sin(t )   0
s 0
 E ( s) cos(t   s)ds (105)

Using the trigonometric identity cos(   )  cos  cos   sin  sin  we obtain
 (t )  E 0 sin(t )
s  (106)
 0  E ( s )  cos(t ) cos( s)  sin(t ) sin( s)  ds
s 0

which gives after some rearrangement

© 1999–2021 Eyal Levenberg 25


Pavement Mechanics (11451) August 2021

 s 

 (t )   0  E  


s 0
E ( s ) sin( s )ds  sin(t )

în-phase with appliedstrain
(107)
 s 

  0    E ( s ) cos( s )ds  cos(t )
 s 0 
90°out-of-phase with appliedstrain

Comparing with previous derivation gives


s 
E1 ( )  E   
s 0
E (s)sin( s)ds
(108)
s 
E2 ( )   
s 0
E ( s) cos( s)ds
t 
E1 ( )  E     E(t )  E  sin(t )dt
t 0

(109)
t 
E2 ( )     E(t )  E  cos(t )dt
t 0

These expressions converge because of the requirement that lims E(s)  0 . They provide

a direct mathematical relation between the frequency-domain dynamic modulus (and its storage

and loss components) and the time-domain relaxation modulus.

5.2 The complex modulus

Because E1 and E2 are essentially normal entities (90° out-of-phase), a complex representation

suggests itself
E*  E1  iE2 (110)

In which E * is the linear viscoelastic 'complex modulus,' having the 'storage modulus,' E1 , as

its real component and having the 'loss modulus,' E2 as its imaginary component.

Using complex notation, the applied strain history is expressed as


 (t )   0eit   0  cos(t )  i sin(t )  (111)

and the resulting steady-state stress is


 (t )   0ei (t  )   0eit ei   0  cos(t   )  i sin(t   )  (112)

In an Argand diagram the stress and strain can be visualized as two rotating vectors with an
identical angular frequency  . The stress leads the strain by an angle  which changes

© 1999–2021 Eyal Levenberg 26


Pavement Mechanics (11451) August 2021

depending on the level of  . The length of the strain vector remains constant at  0 while the
length of the stress vector varies with  .

Im
 (t )


 (t )
Re

Figure 16. Argand diagram of sinusoidal stress and (lagging) strain.

Both cosine and sine excitations are included is the complex strain representation. Since we

deal with steady-state conditions, either the real or imaginary parts can be viewed as physically

meaningful. Regardless of our observation choice, the complex modulus expressing the ratio

of stress and strain is unique, directly obtained by the quotient

 (t )  0ei (t  )  0 i  0
E*    e   cos( )  i sin( )   E1  iE2 (113)
 (t )  0eit 0 0
To formally arrive at the above derivation, we begin by substitution of the sinusoidal strain

history, i.e., Euler's formula, into the hereditary integral (with d ( )  i 0ei d )
 t


it
 (t )  E 0e  i 0 E (t   )ei d (114)


As before, changing the integration variable from  to s using t   s , from which we have
ds  d , generates a corresponding change of integration limits: s   (  ) and

s  0(  t ) , and yields


s 


it
 (t )  E 0e  i 0 E (s)ei (t s ) ds (115)
s 0
s 
 (t )  E 0eit  i 0eit 
s 0
E (s)eis ds (116)

 s 

 (t )   0 eit  E  i


s 0
E ( s )e i s ds 

(117)

© 1999–2021 Eyal Levenberg 27


Pavement Mechanics (11451) August 2021

 s 

 (t )   0eit  E  i


s 0
E ( s )(cos( s )  i sin( s ))ds 

(118)

 s  s 

 (t )   0eit  E  


s 0
E ( s ) sin( s )ds  i 
s 0
E ( s ) cos( s )ds 

(119)

 (t )   0eit  E1  iE2  (120)


 0 0 
 (t )   0eit  cos( )  i
sin( )  (121)
 0 0 
 (t )   0e  cos( )  i sin( )    0e e
it it i
(122)

Additional relations of interest


lim E1 ()  lim E(t )  E0 and lim E2 ( )  0 (123)
  t 0  

The complex compliance is defined as: D*  1/ E*  D1  iD2 , which under steady-state

sinusoidal excitation gives:  (t )  D * (t ) . Also: tan( )  D2 / D1  E2 / E1 .

Assignment #4: Complex Modulus Relations

© 1999–2021 Eyal Levenberg 28


Pavement Mechanics (11451) August 2021

6. Time-Temperature Superposition

The time domain material properties D(t ) and E (t ) and also the frequency domain properties

D *( ) and E *() are applicable for a given reference/test temperature, T0 . The assumption

of thermo-rheological simplicity states that these relations can remain applicable for a different

(albeit constant) temperature, T , simply by replacing physical time with reduced (or pseudo)

time, t r , defined as tr  t / aT , or by replacing the angular frequency with reduced angular

frequency, r , defined as r   aT . In both cases, aT  aT (T , T0 ) is the so-called time-

temperature shift factor which is a function of temperature only.

For the selected reference temperature, i.e., T  T0 , the time-temperature shift factor
equals unity by definition. For the case in which T  T0 , aT decreases to a value smaller than

unity ( aT is always positive) and for T  T0 , aT increases (increase is unrestricted). Hence,

frequency (or time) domain material functions obtained at different temperatures can be viewed

as measured under one (reference) temperature, T0 , but at different frequencies (or times).

Using this feature, the time-temperature shift factor aT provides means for extending the

experimental data range beyond that for which direct measurement methods are available.

For instance, when plotted with respect to the reference temperature, T0 , complex

modulus test data obtained at T  T0 will be translated horizontally to low frequencies.

Similarly, relaxation modulus information obtained at T  T0 will be translated horizontally

along the positive direction on the time axis. These translations form so-called "master curves"

while at the same time they help define the time-temperature shift factor.

For a certain class of polymers, and for a limited range of temperatures, aT tends to follow the
WLF equation, which is based on the assumption that aT (T , T0 )  aT (T  T0 )
c1 (T  T0 )
log( aT )  (124)
c2  (T  T0 )

in which c1 and c 2 are both positive constants; c1 is unitless and c 2 has temperature units.

This equation was found applicable to bituminous materials, both binders and asphalt concrete.

© 1999–2021 Eyal Levenberg 29


Pavement Mechanics (11451) August 2021

Figure 17. Asphalt concrete master curve in the frequency domain.


log E (t )

log t
Figure 18. Asphalt concrete master curve in the time domain.

Project #4: Mastercurve Construction

© 1999–2021 Eyal Levenberg 30


Pavement Mechanics (11451) August 2021

7. Mechanical Analogs and Spectra

In the early development of the theory of viscoelasticity, material behavior was

diagrammatically represented using a collection of massless elastic and viscous elements. Such

modeling reveals nothing about the structural details of the material being deformed; it is

merely useful for illustrating the physical significance of a constitutive equation and

approximating observed behavior for numerical calculations.

The elastic element is a Hookean spring characterized by a stiffness constant E (stress


units) and obeying the relation  (t )  E (t ) . The viscous element is a Newtonian dashpot

characterized by viscosity constant  (units of stress×time) and obeying the relation

 (t )   (t ) .

Hookean Newtonian
spring 
l 1 E l 1 dashpot




Figure 19. Basic mechanical analog elements.

Class Exercise #9: Dashpot under Harmonic Loading

Hookean springs and dashpots can be combined in many different ways to describe

viscoelastic bodies. The two most basic models are Maxwell (James Clerk Maxwell 1831-

1879) and Kelvin-Voigt (Lord Kelvin 1824-1907; Woldemar Voigt 1850-1919).

The Maxwell model consists of a spring and dashpot arranged in series, reproducing

fluid-like behavior. The operator equation for the Maxwell model is obtained by noting that

the overall stress at any point in time  (t ) equals the stress in the spring  E (t ) and also equals
that in the dashpot  (t ) (i.e.,    E   ). At the same time, the overall strain  (t ) equals

the sum of the individual strains (i.e.,    E   ). Consequently, the constitutive equation for
the Maxwell model is

© 1999–2021 Eyal Levenberg 31


Pavement Mechanics (11451) August 2021

1 1 d (t ) d  (t )
 (t )   (125)
m Em dt dt

Em Maxwell

l 1

m


Figure 20. Maxwell mechanical analog.

The Kelvin-Voigt model consists of a spring and a dashpot arranged in parallel.

Kelvin-Voigt

Ekv kv

Figure 21. Kelvin-Voigt mechanical analog.

The constitutive equation for the Kelvin-Voigt model is obtained by noting that the overall

strain  (t ) equals the strain in the spring  E (t ) and also equals that in the dashpot  (t ) (i.e.,

   E   ). The overall stress  (t ) equals the sum of the individual stresses (i.e.,
   E   ). Consequently, the governing equation for the Kelvin-Voigt model is
d  (t )
 (t )  Ekv (t )   kv (126)
dt

© 1999–2021 Eyal Levenberg 32


Pavement Mechanics (11451) August 2021

To explore the behavior of the Maxwell model, a virtual relaxation experiment is carried out

by "applying" a step-strain history of the form  (t )   0 H (t ) . The resulting stress relaxes


according to the expression
 (t )  Em 0et /  m (127)

in which m  m / Em (units of time) is a characteristic "relaxation time" controlling the rate

in which the stress relaxes. Physically,  m represents the time needed for the stress to fall to

e 1 of its initial value; it is also the time at which the stress function passes through an inflection

when plotted against log time. By normalizing the above expression with respect to  0 the

relaxation modulus of the Maxwell model is obtained


E (t )  Emet / m (128)

When a Kelvin-Voigt model is exposed to a virtual creep experiment, i.e., a suddenly applied

stress history of the form  (t )   0 H (t ) , the resulting strain is


0
 (t ) 
Ekv
1  e  t / kv
 (129)

in which  kv  kv / Ekv (units of time ) is a characteristic "retardation time" controlling the

development of delayed/retarded deformation.

By normalizing the above expression with respect to  0 the creep compliance for the Kelvin-

Voigt model is obtained


D(t )  Dkv 1  et / kv  (130)
in which Dkv  1/ Ekv .

Assignment #5: Mechanical Viscoelastic Analogs

More complexity is needed to reproduce realistic viscoelastic behavior. This can be

achieved by combining several basic Maxwell or Kelvin-Voigt elements. Any number of

Maxwell elements in series or any number of Kelvin-Voigt elements in parallel can be reduced

to a single Maxwell or Kelvin-Voigt element (respectively). However, Maxwell elements in

parallel or Kelvin-Voigt elements in series have more complicated properties. In the case of
the Generalized Maxwell model ( N elements in parallel), the relaxation modulus is

© 1999–2021 Eyal Levenberg 33


Pavement Mechanics (11451) August 2021

N
E (t )   Ei e  t / i (131)
i 1

with i  i / Ei .

To reproduce solid behavior, one of the dashpots must have infinite viscosity (or

equivalently, one of the relaxation times must be infinite), and the corresponding spring

constant becomes the equilibrium modulus E . This is called Maxwell-Weichert model:

E1 E2 Ei

E

1 2 i

Figure 22. Generalized Maxwell-Weichert mechanical analog for modeling a linear


viscoelastic solid.

A Generalized Kelvin-Voigt model ( M elements in series) is shown below:

E1 E2 EM
E0

1 2 M

Figure 23. Generalized Kelvin-Voigt mechanical analog for modeling a linear viscoelastic
solid.

© 1999–2021 Eyal Levenberg 34


Pavement Mechanics (11451) August 2021

The creep compliance of the generalized Kelvin-Voigt mechanical analog is

 
M
D (t )   Di 1  e  t / i (132)
i 1

with Di  1 / Ei and  i  i / Ei . To account for an instantaneous compliance (i.e., elastic

response upon loading), one of the dashpots must have zero viscosity (or one of the retardation

times must be zero), and the corresponding spring constant is denoted by D0 (  1/ E0 ).

Mathematically, the constitutive equations for the generalized models may be cast in the

following differential form

 d d2   d d2 
 0
p  p1  p2 2
 ...  (t )  q 
 0 1 q  q2 2
 ...  (t ) (133)
 dt dt   dt dt 
in which the pi 's and qi 's are constants representing material properties. If chosen

appropriately, the generalized models (Maxwell-Weichert and Kevin-Voigt) can be made

mathematically equivalent and thus depict identical behavior.

The generalized models have discrete spectrums of relaxation times and retardation

times: each relaxation time  i is associated with a spectral strength Ei and each retardation

time  i is associated with a spectral compliance magnitude Di . In principle, the above models

may be used to fit relaxation or creep test data by taking N or M to be sufficiently large.

However, the determination of the discrete spectrum lines (times and intensities) in these cases

is not unique.

If the number of elements in the generalized models is increased without limit, the result

will yield continuous relaxation and retardation spectrums which are unique

 h( )e
 t /
E (t )  E  d ln  (134)
 0

 l ( ) 1  e  d ln 
 t /
D(t )  D0  (135)
0

in which  is a time-like integration variable. h( ) is the relaxation spectrum (units of stress),

l ( ) is the retardation spectrum (units of stress-1), with E and D0 previously defined.

Because experience has shown that logarithmic time scales are more convenient, the integration

© 1999–2021 Eyal Levenberg 35


Pavement Mechanics (11451) August 2021

is carried out with respect to ln  and not with respect to  . The functions h( ) and l ( ) are
interrelated (e.g., Gross and Pelzer, 1952).

Of all the viscoelastic functions, the relaxation spectrum (or equivalently the retardation

spectrum) is considered to be the most closely related to the underlying chemo-physical

mechanisms responsible for the viscoelastic behavior. Two possible candidates for asphalt

concrete are
h( )  a1 (a3 /  )a2 exp(a3 /  ) (136)
h( )  a1 exp(a2 (ln( / a3 ))2 ) (137)
In both cases, a1 has stress units, a3 has units of time, and a2 in unitless. The components of
the complex modulus can be expressed in terms of the spectra functions

 2 2
E1 ( )  E   h( ) d ln  (138)
 0 1   2 2


E2 ( ) 

 h( ) 1   
0
2 2
d ln  (139)

© 1999–2021 Eyal Levenberg 36


Pavement Mechanics (11451) August 2021

Appendix: Bibliography

Burmister, D. M. (1945), "The General Theory of Stresses and Displacements in Layered Systems,"
Journal of Applied Physics, Vol. 16, pp. 89-94 (Part I); pp. 126-127 (Part II); pp. 296-302 (Part III).
Burmister, D.M. (1943), "The Theory of Stresses and Displacements in Layered Systems and
Application to the Design of Airport Runways," Proceedings of the Highway Research Board, Vol. 23,
Washington, D.C., pp. 126–148.
Christensen, R. M. (2003), "Theory of Viscoelasticity," Second Edition, Dover, New York, pp. 378.
Ferry, J. D. (1980), "Viscoelastic Properties of Polymers," Third Edition, John Wiley & Sons, Inc., New
York, pp. 641.
Flügge, W. (1975), “Viscoelasticity,” Second Edition, Springer Verlag, Berlin, pp. 194.
Gross B. and Pelzer H. (1952), "Relations between Delta Functions," Proceedings of the Royal Society
of London. Series A, Mathematical and Physical Sciences, Vol. 210, No. 1102, pp. 434–437.
Haddad, Y. M. (1995), "Viscoelasticity of Engineering Materials," Chapman & Hall, London.
Hopkins, I. L. and Hamming, R. W. (1957), "On Creep and Relaxation," Journal of Applied Physics,
Vol. 28, No. 8, pp. 906-909.
Huang, Y. H. (2004), "Pavement Analysis and Design," 2nd Edition, Pearson Prentice Hall.
Lockett, F. J. (1972), "Nonlinear Viscoelastic Solids," Academic Press.
Longman, I. M. (1956), "Note on a method for Computing Infinite Integrals of oscillatory Functions,"
Cambridge Philosophical Society Proceedings, Vol. 52, p. 764.
Lytton, R. L. (1989), "Backcalculation of Pavement Layer Properties," Nondestructive Testing of
Pavements and Backcalculation of Moduli, ASTM STP 1026, A.J. Bush III and G. Y. Baladi, Eds.,
American Society for Testing and Materials, Philadelphia, pp. 7–38.
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, Vol. 36, pp. 1653–1675.
Schapery, R. A. (1962), "Approximate Methods of Transform Inversion for Viscoelastic Stress
Analysis," ASME Proceedings of the 4th U.S. National Congress on Applied Mechanics Vol. 2, 1075–
1085.
Schwarzl, F. and Staverman, A. J. (1952), "Time-Temperature Dependence of Linear Viscoelastic
Behavior," Journal of Applied Physics, Vol. 23, No. 8, pp. 838-843.
Soo, C. K. (2009), "Geometric Interpretation of Time-Temperature Superposition," Korea-Australia
Rheology Journal, Vol. 21, No. 1, pp. 13–16.
Sugihara, M. (1987), "Methods of Numerical Integration of Oscillatory Functions by the DE-formula
with Richardson Extrapolation," Journal of Computational and Applied Mathematics, Vol. 17, pp. 47–
68.
Uzan, J. (1994), "Advanced Backcalculation Techniques," Nondestructive Testing of Pavements and
Backcalculation of Moduli (second volume), ASTM STP 1198, H. L. Von Quintus, A. J. Buch III, and
G. Y. Baladi, Eds., American Society for Testing and Materials, Philadelphia, pp. 3–37.
Uzan, J., Lytton, R. L., and Germann, F. P. (1989), "General Procedure for Backcalculating Layer
Moduli," Nondestructive Testing of Pavements and Backcalculation of Moduli, ASTM STP 1026, A.J.
Bush III and G. Y. Baladi, Eds., American Society for Testing and Materials, Philadelphia, pp. 217–
228.
Van Cauwelaert, F. (2003), "Pavement Design and Evaluation—The Required Mathematics and its
Applications," M. Stet (ed). Federation of the Belgian Cement Industry, Brussels, Belgium.

© 1999–2021 Eyal Levenberg 37

You might also like