Professional Documents
Culture Documents
A Theory Relating Creep and Relaxation For Linear Materials With Memory
A Theory Relating Creep and Relaxation For Linear Materials With Memory
1 Introduction convolution theorem, e.g., Gurtin and Sternberg 共see Theorem 1.2
共g兲 in Ref. 关4兴兲. The Stieltjes convolution theorem states that if
Volterra 关1,2兴, by utilizing the theory of integral equations, was
f共t兲 and g共t兲 are equal to zero in the region t 苸 共−⬁ , 0兲 and f and
a major contributor to the theory of viscoelasticity. The Riesz
representation theorem for a linear functional may be used to ob- g共1兲 are continuous in the interval t 苸 关0 , ⬁兲, then
tain the basic constitutive equations for linear materials with f ⴱ dg = g共0兲f + f ⴱ g共1兲 共1.2兲
memory. They are given by
Thus, we have
冕 冕
t
t
= G ⴱ d = G共t − 兲d共兲 and d
共t兲 = G共t兲共0兲 + G共t − 兲 d and
0
0
d
冕 冕
t t
= J ⴱ d = J共t − 兲d共兲 共1.1兲 d
共t兲 = J共t兲共0兲 + J共t − 兲 d 共1.3兲
0 0
d
where the Stieltjes convolution integral is denoted by *, G and J Let ij and sij be the Cartesian components of the symmetric
denote the relaxation modulus and creep compliance, respectively, stress tensor and the components of the deviatoric stress tensor,
is the stress, and is the strain. J is the Stieltjes inverse of G, respectively. If ij are the Cartesian components of the infinitesi-
and conversely, G is the Stieltjes inverse of J. The basic constitu- mal strain tensor and eij are the deviatoric components of the
tive equations can be put in their customary form by utilizing the infinitesimal strain tensor, then the stress strain relations for an
properties of the Stieltjes integral, i.e., functional analysis by isotropic linear Viscoelastic material are sij = 2 ⴱ deij and kk
Riesz and Sz.-Nagy 共see pp.105–140 in Ref. 关3兴兲 and the Stieltjes = 3k ⴱ dkk. The two functions 共t兲 and k共t兲 are the isotropic relax-
ation functions in shear and dilatation, respectively.
Rabotnov 关5兴 presented a general theory of hereditary solid me-
Contributed by the Applied Mechanics Division of ASME for publication in the
JOURNAL OF APPLIED MECHANICS. Manuscript received August 31, 2008; final manu-
chanics using integral equations, and Koeller 关6兴 reviewed the use
script received August 27, 2009; published online February 8, 2010. Assoc. Editor: of integral equations for viscoelasticity and interjects fractional
Krishna Garikipati. calculus into Rabotnov’s theory by the introduction of the spring-
Journal of Applied Mechanics Copyright © 2010 by ASME MAY 2010, Vol. 77 / 031008-1
Downloaded 28 May 2010 to 137.104.114.22. Redistribution subject to ASME license or copyright; see http://www.asme.org/terms/Terms_Use.cfm
pot, which he used to generalize the classical models. Meshkov et number of overlapping pairs of sign changes in the curvature.
al. 关7兴 as well as Rossikhin and Shitikova 关8兴 summarized Rabo- Researchers, whose primary interest is in creep, comprehend that
tnov’s theory. Rabotnov’s fractional exponential function is re- nonlinear effects and yielding must be incorporated.
lated to the well known Mittag-Leffler function and they showed Mechanical models of springs and dashpots are frequently used
the equivalence of Rabotnov’s theory to Torvik and Bagley’s frac- to obtain the relaxation modulus and corresponding creep compli-
tional polynomial constitutive equation. Further references on ance. Flügge 关29兴 presented an introduction to viscoelastic models
Rabotnov’s theory may be obtained in Ref. 关8兴. Mainardi and using the differential equation space. In Table 1 共see Ref. 关29兴兲, he
Gorenflo 关9兴 gave a review of relaxation processes using frac- gave the model type, the name of the model, and the differential
tional calculus in their tutorial survey. There are many applica- equation describing that model; in addition, he gave the creep
tions of fractional calculus in viscoelasticity. Fraggstedt 关10兴 used compliance and relaxation modulus as well as the complex com-
a fractional Kelvin–Voigt model to analyze the power dissipation pliance for a set of elemental models. Gurtin and Sternberg 关4兴
in tires. The fractional calculus Maxwell model for rest state sta- presented an extensive article on the polynomial differential con-
bility has been used by Heibig and Palade 关11兴. Beris and Ed- stitutive equations when the powers are of integer order. The
wards 关12兴 showed that models with spring-pots and their frac- present method is to combine these various approaches into a
tional calculus differential equations satisfy the thermodynamic geometrical presentation by comparing the relationship between
admissible criteria. Heymans 关13兴 introduced the spring-pot into the creep compliance and relaxation modulus. It will be done in
hierarchical models and presented experimental data for the three- the integral-differential equation space rather than in the function
stage Zener model having ten parameters. She did not obtain any space or the differential equation space. An integral-differential
of the viscoelastic functions, but used a recursive procedure to equation will relate the various methods both empirically and ana-
obtain these ten parameters. Heymans used experimental data lytically for the development of creep and relaxation functions for
from Read 关14兴 for isotactic polypropylene to determine the ten linear materials with memory. When the memory parameter beta
parameters. A Kelvin–Voigt solid with a fractional dashpot and a is one, several plots for the various elementary models is given in
nonlinear spring was used to model Form 1 polyvinylidene fluo- Ref. 关6兴. From these graphs, it can be seen that to represent vis-
ride 共PVDF兲 by Sathiyanarayanan et al. 关15兴. Dîkmen 关16兴 used coelastic materials in creep and relaxation requires a model of the
the fractional calculus Kelvin–Voigt model in modeling seismic standard linear solid, also called the three parameter solid model
waves in buildings. A stochastic dynamic system, when the damp- or the Zener model 关30兴. This observation may also be deduced
ing is selected as a half-order derivative, was used by Agrawal from Eq. 共2.1兲 and its corresponding creep compliance and relax-
关17兴. Rossikhin and Shitikova 关18兴 considered a damped oscillator ation modulus. The standard linear solid gives only a rudimentary
involving a fractional parameter and obtained an analytical solu- description of the behavior of the viscoelastic material. The dif-
tion for the problem. ferential equation of the standard linear solid is
再 冎
Many research papers have been published on the fractional
diffusion-wave equation of Mainardi 关19,20兴. In other research d共t兲 d共t兲
fields interesting applications of the fractional calculus occur in + p0共t兲 = E0 + q0共t兲 共2.1兲
dt dt
automatic control and signal processing 关21,22兴, as well as, the
Fokker–Planck equation, which describes anomalous diffusion Its relaxation modulus is G共t兲 = E0兵exp共−t / p0−1兲 + q0 / p0关1
and relaxation processes 关23,24兴. The spring-pot concept has been
− exp共−t / p0−1兲兴其 and its creep compliance is J共t兲
applied to describe the arterial viscoelasticity by Craiem and Ar-
mentano 关25兴. = E0−1兵exp共−t / q0−1兲 + p0 / q0关1 − exp共−t / q0−1兲兴其.
The relaxation modulus looks like a rounded downward double
2 Geometrical Interpretation of the Half-Order Con- staircase as developed by Tobolsky and Catsiff 关31兴. From Figs. 4
and 5 in Ref. 关6兴, we see that for a first-order transition when beta
stitutive Equation is one, the standard linear solid has an elastic response and a
Much of the pioneering works on viscoelasticity use empirical transition region as well as a limiting value as time approaches
equations and they are still useful in reducing experimental data. infinity. The corresponding creep compliance has an initial re-
In relaxation, one often-used empirical equation is time raised to a sponse, and primary and secondary creeps; however, the response
power, which is called the Nutting power law 关26兴. Catsiff and does not have a long enough time domain required for fitting
Tobolsky 关27兴 used the Gauss error function in their reduction for experimental data. With the introduction of the memory parameter
the relaxation data in the transition region for National Bureau of beta the time scale is stretched out, which gives this model a
Standards 共NBS兲 Polyisobutylene. There are many empirical better representation to account for experimental data 共see Refs.
equations used in creep data reduction. In the primary region, time 关32–36兴, all of which have used a first-order transition model兲.
raised to a power is often used. Rabotnov 关28兴 gave some of these Scott Blair 关37兴 used fractional calculus in his atypical research in
proposed creep empirical equations. the dairy industry, but his equation gave the basic units in frac-
Several analytical approaches have been used to determine the tional powers, which were one of the various criticisms of his 17
unknown relaxation modulus denoted by G共t兲 and the unknown published papers.
creep compliance J共t兲. In reducing experimental data for creep We review again the second-order transition but focus on the
and relaxation, three different regions are generally specified. For geometrical shapes of the creep compliance and relaxation modu-
creep the classifications are the initial response, the primary creep lus. We specifically consider their reflection and distortion as dis-
region, the secondary creep region, and the tertiary region. A com- played in the graphs. The power law of Nutting is a straight line
parable creep classification is the instantaneous extension during on the Log-Log graph of the relaxation modulus, but it does not
the application of the load, the transient part, the steady-state part, have the general rounded characteristic of the standard linear
and the accelerated part that leads to fracture. The corresponding solid. Experimental data usually observe the rounded downward
relaxation classifications are the glassy region, the transition re- single staircase; hence, the power law will not realistically repre-
gion, and quasistatic rubbery plateau region 共maybe the liquid sent some of the experimental data as a “single-plateau.” For a
region兲. The common thing that separates these various regions is linear theory a negative slope on the Log-Log plot of the relax-
that there is a change in the sign of the curvature that occurs in ation modulus is needed to represent the transition region so that
both the creep and relaxation curves. When there are overlapping the power of the Nutting’s equation is between zero and a nega-
pairs of sign changes in the curvature in the relaxation modulus or tive one. Thus, we take the Nutting’s equation proportional to t−,
creep compliance, the theory to describe the material behavior where  苸 共0 , 1兲. In order to combine the relaxation modulus of a
will be said to undergo a transition. It will be called a transition of standard linear solid and Nutting’s equation into a single expres-
the first kind, a second kind, and a third kind, depending on the sion, we seek an integral expression of Nutting’s equation for this
Downloaded 28 May 2010 to 137.104.114.22. Redistribution subject to ASME license or copyright; see http://www.asme.org/terms/Terms_Use.cfm
particular case. This is accomplished by considering Able’s inte- greater transitions. Pioneering researchers have mentioned that
gral equation. The kernel of the Able’s operator,1 denoted by I, is when we suppose the future of a system at the current time de-
pends on the actual state and near neighboring states which pre-
t−h共t兲 ceded it, then this is a restrictive hypothesis, and they introduce
I共t兲 = , 0ⱕⱕ1 共2.2兲
⌫共1 − 兲 both integration and differentiation together.
where ⌫ denotes the Gamma function and h is the Heaviside unit
step function. If the material does not age, it is wise to select a
number for beta in order that the resulting relaxation modulus or 3 Generalized Functions Approach to Fractional Cal-
creep compliance does not depend on the length of time that the culus
experimental data were collected. Beta is selected to be one-half,
which will result in the empirical expression that was used by It is not the purpose of this paper to give any review of frac-
Catsiff and Tobolsky 关27兴. In order to combine the power law with tional calculus since Podlubny 关40兴 has a book on fractional dif-
the standard linear solid, we select the kernel of Able’s operator, ferential equations along with many references. Other books on
i.e., I1/2共t兲 = t−1/2 / 冑 , as a Stieltjes convolution. It is put in both fractional calculus are by Oldham and Spanier 关41兴, and Ross and
sides of the standard linear solid model differential equation but co-worker 关42,43兴. There are books on linear operators, in which
with different coefficients. One may set p1/2, q1/2, or both equal to integro-differential equation are given for hereditary mechanics,
zero, but we introduce terms to both sides of Eq. 共2.1兲 so that the even as early as 1936.
creep compliance and the relaxation modulus have some type of The present approach is to use the concept of the primitive of
mirror, reflected, and distorted images. Thus, this second-order order of the function g as a convolution. The primitive of order
transition model is an integral-differential equation given by was introduced by Gel’fand and Shilov 共see Ref. 关44兴, pp. 115–
122兴. It is also discussed in pp. 81–86 in Ref. 关40兴. We use the
冕
t
d共t兲 p1/2 notation of Gel’fand and Shilov for generalized functions only in
+ 共t − 兲−1/2d共兲 + p0共t兲 this section. They defined a function x+ equal to x 共for x ⬎ 0兲,
dt 冑
再 冎
0
and zero 共for x ⱕ 0兲 so that it is of bounded support and can be
冕
t described by a regular functional. They introduced a generalized
d共t兲 q1/2
= E0 + 共t − 兲−1/2d共兲 + q0共t兲 共2.3兲 function ⌽ by the expression x+−1 / ⌫共兲 , which remains con-
dt 冑 0 centrated on the half-line x ⱖ 0. The generalized function x+ was
We may look at this development as a combination of the standard normalized with e−x so that the simple poles of x+ are eliminated.
linear solid and Nutting’s equation not in the algebraic space nor An important property of the generalized function ⌽ is
in the differential space but in the differential integral space. The
square root of pi is being introduced so that t−1/2 / 冑 is normal- ⌽−k = ␦共k兲共x兲 for k = 0,1,2. . . 共3.1兲
ized. Taking the Laplace transformation of Eq. 共2.3兲 when the
initial conditions are zero results in where ␦共k兲共x兲 is the kth derivative of the generalized delta func-
tional. Then, they defined the primitive of order of the function
共s + p1/2s1/2 + p0兲L 兵共t兲其 = E0共s + q1/2s1/2 + q0兲L 兵共t兲其 g as the convolution g = g共x兲 * ⌽, where * now denotes the con-
共2.4兲 volution of generalized functions 共see Ref. 关44兴, pp. 103–106兲. By
considering the case when is negative, they defined the deriva-
where L denoted the Laplace transform operator. By a change in
variables in Eq. 共2.4兲 we see that this is a second-order transition. tive of order of the function g as
The relaxation and creep functions corresponding to Eq. 共2.4兲 are
given in Ref. 关38兴. One may use initial conditions in taking the dg/dx = g− = g ⴱ ⌽− 共3.2兲
Laplace transform of Eq. 共2.3兲 using the method of Heymans and In order to relate this theory to the notation used in fractional
Podlubny 关39兴 for initial condition in fractional differential equa- calculus and to show the correlation of the theorem in Ref. 关45兴
tions. Researchers that are looking for equations to fit the experi- 共the jump term in the Stieltjes convolution was overlooked兲, we
mental data need not be concerned with fractional calculus. Frac- introduce the operational notation D and D−1 to denote differen-
tional calculus history is nearly 300 years old and it is discussed in tiation and integration, respectively, that is, Dg = dg / dx and
Sec. 3. Catsiff and Tobolsky’s 关27兴 experimental data for the D−1g = 兰x0 g共u兲du. Now operators D and D−1, in general, do not
glassy and leathery rubbery regions using the derived relaxation commute unless g共0兲 = 0 since DD−1g = D−1Dg + g共0兲. It is impor-
function given in Ref. 关38兴 were used and it gave a good compari- tant, but not essential, to put the terminal on the operator and write
son for their experimental data 共note that in Ref. 关38兴 a negative Dx for differentiation with respect to x and 0Dx−1 for integration
sign is omitted in the definition of the Riesz distribution, and in
between the limit 0 and x. For fractional differentiation Dx is
the denominator of the loss tangent, the positive sign should be
negative兲. A third order transition is specified in Sec. 4 that ac- used and for fractional integration 0Dx− is used. It then follows
counts for the flow region in the relaxation modulus as well as the that Gel’fand and Shilov’s theory may be written as 0Dx−g
tertiary region in the creep compliance. It is very unlikely that for = g共x兲 ⴱ ⌽ for fractional integration and Dxg = g ⴱ ⌽− for frac-
viscoelastic materials, more than a third order transition will be tional differentiation. In Refs. 关45,38兴 the Riesz distribution was
required to fit the experimental data. Even though this third order introduced, which is in terms of the present notation given by
transition model is a solid model when a certain creep time is
selected in a special way, the model becomes a fluid. Since the R共x兲 = 0 for xⱕ0 and for
relaxation and creep functions will be given in an analytical form,
one will be able to determine if the time-temperature superposi-
x ⬎ 0 it is R共x兲 = x+−/⌫共1 − 兲 共3.3兲
tion principle can be used. Thus, this theory should have interest-
ing applications for a large class of materials such as steel at The relationship between the Riesz distribution and the
temperatures much higher than room temperature, for some bio- Gel’fand and Shilov generalized function ⌽ is ⌽ = R1−. Substi-
logical materials that grow and then decay may require still tution of ⌽ = R1− into 0Dx−g = g ⴱ ⌽ gives 0Dx−g = g ⴱ R1−.
Also for fractional differentiation, we have Dxg = g ⴱ R1+. One of
1 the properties of the Riesz distribution is when m is a positive or
If beta is both positive and negative, then the Able’s operator considered as a
generalized functional becomes the fractional integral and fractional derivative 共see negative integer then DmR = R+m. This may be obtained as fol-
Sec. 3兲. lows. From p. 24 of Ref. 关44兴 they show Dx+ = x+−1. Thus
Downloaded 28 May 2010 to 137.104.114.22. Redistribution subject to ASME license or copyright; see http://www.asme.org/terms/Terms_Use.cfm
DR共x兲 = D关x+−/⌫共1 − 兲兴 = − x+−−1/⌫共1 − 兲 = x+−共1+兲/⌫共− 兲
= R1+共x兲 共3.4兲
and by a recursive process, we obtain the result that DmR
= R+m. The expression for fractional differentiation in terms of
the Riesz distribution is
Fig. 1 The mechanical model for a third order transition
Dxg = g ⴱ R1+ = g ⴱ DR = R ⴱ Dg = R ⴱ dg 共3.5兲
In the calculation of Eq. 共3.5兲 the use of the convolutions of gen-
eralized functions has been made 共see p. 105 in Ref. 关44兴兲, that is, Kelvin, Maxwell, Voigt, Zener, and others have introduced me-
D共f * g兲 = Df * g = f * Dg as well as f * g = g * f. Also dg chanical models to try to represent Viscoelastic materials. They
= 共dg / dx兲dx. By a similar procedure we can write the Gel’fand have some usefulness since they give a small insight into the
and Shilov equation for fractional integration in terms of the Riesz behavior of the material. The model in Fig. 1 is called the me-
distribution as chanical model. It is called the mechanical model because of all
the numerous models it has the ability to represent and maybe
0D x
−
g = R ⴱ dg 共3.6兲
predict the extended time response of linear viscoelastic materials
Or as Gel’fand and Shilov did, replace lambda by a negative with memory. It considers the reflection and distortion of the ex-
lambda. What transpired under this change in notation is that the perimentally observed creep compliance and relaxation modulus.
fractional integration and fractional differentiation have a Riesz For the mechanical model the dashpot in what is generally called
representation. Note the similarity between Eq. 共1.1兲 and Eq. the Kelvin–Voigt model is replaced by the spring-pot. The result-
共3.5兲. ing creep compliance and relaxation modulus are referred to as the
Let us consider Eq. 共3.5兲 creep and relaxation functions. The spring-pot element is 共t兲
冕
x = ED共t兲, where E is the modulus of elasticity for the spring
Dxg共x兲 = 共R ⴱ dg兲共x兲 = g共0兲R共x兲 + R共x − 兲g共1兲共兲d element = F / E and F is the coefficient of viscosity for a dashpot
0 element. The characteristic time is called the relaxation time or
the creep time, depending on the model. The spring-pot is a
共3.7兲
Hook’s solid when the memory parameter is zero and a Newton-
Substitution of R共x兲 = x− / ⌫共1 − 兲 results in ian fluid when beta is 1. A spring-pot is a nondimensional form of
冕
x Scott Blair’s equation, but materials do not correspond to this
g共0兲x− 1
Dxg共x兲 = + 共x − 兲−g共1兲共兲d 共3.8兲 equation. The uses of the fractional calculus mechanical models
⌫共1 − 兲 ⌫共1 − 兲 0
were introduced to obtain Rabotnov’s general theory and are
given by Eqs. 共4.9兲 and 共4.10兲 in Ref. 关6兴. In order to obtain an
This is equation 共1.9兲 in Ref. 关9兴 when 苸 共0 , 1兲. If g共0兲 = 0, then initial strain due to a suddenly applied stress as well as to account
it is the Caputo fractional derivative as obtained in Refs. for wave propagation, a spring was introduced. The fractional
关33,46,9兴, where it is denoted in Ref. 关9兴 by ⴱDx. exponential function was introduced in 1948 by Rabotnov, and he
Furthermore if g共0兲 = 0 in the interval x 苸 共−⬁ , 0兴 then g共x兲 is showed the relationships to the Mittag-Leffler function. The
zero up to and including zero, so that all its derivatives are zero at Mittag-Leffler function 关47兴 was defined in 1905 and is available
time equal to zero. For x ⬎ 0 it is sufficient, for many applications, on some commercial programs.
to assume that g共x兲 is continuous and bounded. In that case D The mechanical model of this theory for a third order transition
commutes with its inverse and DD = DD for all lambda and is a chain of three Kelvin–Voigt fractional calculus models, along
mu, with D0 being the identity operator. That is, we omit the with a spring having a modulus of E0. Figure 1 shows this model,
terminal and simply write Dg = R * dg. When using this opera- where the spring-pot element is displayed like a diamond. The
tional notation it is essential to remember that when is positive third order mechanical model accounts for various transitions that
occur. The first spring represents the instantaneous response or
it is differentiation with respect to x and when is negative it is
wave speed, the first Kelvin–Voigt fractional element accounts for
integration from 0 to x. Also, remember that some functions are
the conformational rearrangement of the polymeric chain resulting
zero up to and including zero so they are of bounded support
in the primary creep, the second Kelvin–Voigt fractional element
while sometimes a function may be in the Heaviside space, e.g.,
accounts for the movement of the side groups within the poly-
h共x兲.
meric chains, resulting in the steady-state part of the creep, and
finally, the third Kelvin–Voigt fractional element accounts for the
polymeric chains beginning to untangle, which leads to the accel-
4 The Mechanical Model for a Third Order Transition erated part and fracture. The material will flow when the creep
time in the third Kelvin–Voigt element tends to infinity.
From now on we use the notation that was used in Secs. 1 and For the mechanical model in Fig. 1 the total stress is the same
2. The current time is denoted by t and the past time by so that in each of the chains as well as the stress in the first spring, while
t − is the elapsed time. D = Dt is the fractional derivative, beta the total strain is the sum of the strains of the chains plus the strain
is positive, and differentiation is with respect to the current time t. in the first spring. Thus, the differential integral equation of state
再 冎
The generalized function R is zero up to and including t = 0 and it for a third order transition in materials with memory is
is of bounded support. It is given by R共t兲 = 0 for t 苸 共−⬁ , 0兴 and 3
for t ⬎ 0 it is R共t兲 = t− / ⌫共1 − 兲. Beta is called the memory pa-
兺
1 1
共t兲 = + 关D + n−兴−1 共t兲 共4.2兲
rameter and R is called the Riesz distribution. Functions that are E0 n=1 Enn
zero up to and including zero are said to be in the hereditary
space. Functions that are zero up to but not including zero are In fitting experimental data the use of different memory param-
members of the Heaviside space. Some of the properties of the eters for each of the three Kelvin–Voigt fractional elements will
Riesz distribution are result in a model having ten parameters. The Zener type ladder
model used in Ref. 关13兴 has ten parameters. The integral equation
R * R = R+, R− * dR = ␦ form of Rabotnov’s theory when the number of Kelvin–Voigt el-
共4.1兲 ements in Fig. 1 is N and each Kelvin–Voigt element has a differ-
Rn共t兲 = ␦共n−1兲共t兲 for n = 1,2,3. . . ent memory n is given in equation 共4.11兲 in Ref. 关6兴. However,
Downloaded 28 May 2010 to 137.104.114.22. Redistribution subject to ASME license or copyright; see http://www.asme.org/terms/Terms_Use.cfm
by selecting only one memory parameter and taking it to be one- parameters and the fundamental set. The fundamental set consists
half, the resulting creep and relaxation functions will contain the of the following parameters associated with the mechanical
error function. For nonaging materials the memory parameter model: memory parameter, instantaneous modulus, three relax-
should not depend on the temperature, the composition of the ation times, and three creep times. This phenomenological funda-
material, or the time duration on which the data has been col- mental set is denoted by S 共) ⫽ S 共E0 ,t1 ,t2 ,t3 , 1 , 2 , 3). The ad-
lected. The dependence on the temperature and composition of the vantage of expressing the behavior of viscoelastic materials in
material will be expressed in terms of the instantaneous modulus terms of the fundamental set S 共) is that E0 is related to the wave
and the creep and relaxation times. For aging materials the speed, and thus, a physical constant. The relaxation times t1, t2,
memory parameter  may be a function of the time of aging. and t3 corresponds to the change in curvature of the relaxation
This third order transition model is an eight parameter model, function, and the retardation or creep times 1, 2, and 3 are
these parameters are , E0, E1, E2, E3, 1, 2, and 3. In Eq. 共4.2兲 related to the changes in curvature of the creep function. The
1, 2, and 3 are the ordered real positive creep times. Equation fundamental set S 共) is a function of the temperature T, with T0
共4.2兲 may be rewritten as a polynomial with eight parameters, being the reference temperature, that is
since p3 and q3 are both equal to one as
3 3 E0 = E0共T兲, tn = tn共T兲, and n = n共T兲 共n = 1,2,3兲
兺
n=0
pnDn共共t兲兲 = E0 兺
n=0
qnDn共共t兲兲 共4.3兲 共4.6兲
where
The relaxation function is a monotonically decreasing function of
time, which follows from Truesdell and Noll’s principle of fading
memory 共see p. 101 of Ref. 关48兴兲. It follows that all the time E0共T0兲 = E0, tn共T0兲 = tn, and n共T0兲 = n
parameters are positive real numbers so that 0 ⱕ t1 ⱕ t2 ⱕ t3 and In addition to the temperature, the fundamental set depends on the
0 ⱕ 1 ⱕ 2 ⱕ 3. Then Eq. 共4.3兲 becomes molecular structure. The Laplace transformation of Eq. 共4.4兲 with
3 3 all initial conditions taken to be zero is given by
兿 共D
n=1
 −
+ tn 兲共t兲 = E0 兿 共D +
n=1
n
−
兲共t兲 共4.4兲
3 3
兺 兺t
zero at t = 0 for this solution of the creep function. Since the stress
p 2 = n−共1 + E0/En兲 = n
−
is in the Heaviside space all of its derivatives are determent from
the expression Dh共t兲 = R共t兲 evaluated at t = 0. Specification of
n=1 n=1
共4.5兲 Using partial fractions and then taking the inverse Laplace trans-
q 3 = 1 formation, one obtains the following creep function
3
3
q 2 = 兺
n=1
n
−
J共t; 兲 = E0−1 − E0−1 兺 J 共兲兵1 − E关− 共t/ 兲兴其
n n
n=1
− − −
= J⬁ + E0 −1
兺 J 共兲兵E关− 共t/ 兲兴其
n n 共4.9兲
q0 = 1 2 3 n=1
The mechanical model gives some insight into the complicated where E0−1 = 1 / E0 is the initial response, J⬁ = E0−1关1 − J1共兲
process taking place in the creep and relaxation. It should be − J2共兲 − J3共兲兴 = E0−1t1−t2−t3− / 兵1−2−3−其 is the maximum
anticipated that there will be changes in curvature of the creep and strain, and J1共兲, J2共兲, and J3共兲 are given by
relaxation functions as the dominance of one Kelvin–Voigt ele-
ment replaces the dominance of another Kelvin–Voigt element as 3
the current time progresses. Further, the orders of coefficients in
this model are same on both sides. Equation 共4.4兲 possesses sym- J1共兲 = − 兵1/共1− − 2−兲共1− − 3−兲其 兿 共t
n=1
n
−
− 1−兲
metry, which was introduced by Koeller 关45兴, so that no prefer-
ence is given to the stress or strain; thus, the number of changes in
the curvature in the creep and relaxation functions are the same, 3
Downloaded 28 May 2010 to 137.104.114.22. Redistribution subject to ASME license or copyright; see http://www.asme.org/terms/Terms_Use.cfm
3
G⬘共兲/E0 = G1⬘共兲G2⬘共兲G3⬘共兲 − G1⬙共兲G2⬙共兲G3⬘共兲
J3共兲 = − 兵3/共3− − 1−兲共3− − 2−兲其 兿 共t
n=1
n
−
− 3−兲
− G1⬙共兲G2⬘共兲G3⬙共兲 − G1⬘共兲G2⬙共兲G3⬙共兲
共4.10兲
G⬙共兲/E0 = G1⬙共兲G2⬘共兲G3⬘共兲 + G1⬘共兲G2⬙共兲G3⬘共兲
As presented above the Mittag-Leffler function 共see Ref. 关47兴 or
pp. 206–209 of Ref. 关49兴兲 is denoted by E and it is + G1⬘共兲G2⬘共兲G3⬙共兲 − G1⬙共兲G2⬙共兲G3⬙共兲
⬁
tn 2 + 共tn− + n−兲共cos /2兲 + tn−n−
E共t兲 = 兺 ⌫共1 + n兲 共4.11兲 where G n⬘共 兲 =
n=0 2 + 2tn−共cos /2兲 + tn−2
By a similar procedure or by interchanging the creep time with the
relaxation time and taking into account E0, we obtain for the 共tn− + n−兲共sin /2兲
relaxation function G共t ; 兲 the following expression: and G ⬙共 兲 = 共4.16兲
2 + 2tn−共cos /2兲 + tn−2
3
When t3 and 3 both tend to infinity, then it becomes the second-
G共t; 兲 = E0 − E0 兺 G 共兲兵1 − E关− 共t/t 兲 兴其
n=1
n n
−
order transition. Also when t3, t2, 3, and 2 all tend to infinity, the
first-order transition is recovered as developed by Caputo and
3
Mainardi 关33兴. Bagley 关51兴, using an approximate quantum me-
= E⬁ + E0 兺 G 共兲兵E关− 共t/t 兲 兴其
n n
−
共4.12兲 chanical description of the molecular energy, developed a first-
n=1 order transition model, which has four parameters. Beside all
In Eq. 共4.12兲 E⬁ = E0关1 − G1共兲 − G2共兲 − G3共兲兴 these limiting cases, one can recover the classical theory when the
memory parameter is one.
= E01−2−3− / t1−t2−t3− is the limit of the relaxation func-
Following Welch et al. 关50兴, we reduce the third order transition
tion as time tends to infinity, and G1共兲, G2共兲, and G3共兲 are
into a second transition for the fluid model. This is accomplished
given by by letting 3 → ⬁ so that E⬁ → 0 and J⬁ → ⬁. Thus, the relaxation
3 function for a second-order transition fluid is
G1共兲 = − 兵t1/共t1− − t2−兲共t1− − t3−兲其 兿 共 n
−
− t 1−兲 3
兺G
n=1
G f 共t; 兲 = E0 nf 共兲兵E关− 共t/tn兲兴其, with
3 n=1
G2共兲 = − 兵t2/共t2− − t3−兲共t2− − t1−兲其 兿 共
n=1
n
−
− t 2−兲
2
3
The subscript f denotes a fluid model and G2f 共兲 as well as
G3共兲 = − 兵t3/共t3− − t1−兲共t3− − t2−兲其 兿 共
n=1
n
−
− t 3−兲
G3f 共兲 are obtained by a cycle change in the indices one, two, and
three with G1f 共兲. Since G3f  = 1 − G1f 共兲 − G2f 共兲, one may ex-
The three inflection points of the creep and relaxation functions press the relaxation modulus of the fluid as
are determined from the second derivative of J共t ; 兲 and G共t ; 兲,
respectively. G f 共t; 兲 = E0兵G1f 共兲关E关− 共t/t1兲兴 − E关− 共t/t3兲兴兴
Welch et al. 关50兴 presented an incisive correlation of the six-
parameter model given in Ref. 关45兴 with various known experi- + G2f 共兲关E关− 共t/t2兲兴 − E关− 共t/t3兲兴兴 + E关− 共t/t3兲兴其
mental data. Taking the limit as both t3 and 3 tends to infinity in 共4.18兲
Eqs. 共4.12兲 and 共4.13兲, the second-order transition for the relax-
ation modulus is obtained. Excellent agreements on many of the When using Eq. 共4.17兲 in a least-squares error or the logarithmic
comparisons were made in Ref. 关50兴. Also they take a limit so that least-square error, one should select the fundamental set
the second-order transition solid model becomes a five-parameter S 共) ⫽ S共E0 ,t1 ,t2 ,t3 , 1 , 2). If one uses Eq. 共4.18兲 one may select
fluid model. For that model, one would not expect very good the seven unknown variables E0, t1, t2, t3,, G2f 共兲, G1f 共兲, and 
results, but it is still impressive in their reduction in all Tobolsky The fractional differential equation for this seven-parameter fluid
and Catsiff’s experimental data. By selecting several memory pa- model is
rameters and using a cutoff function, all of Tobolsky and Catsiff’s
关31兴 experimental data were fitted, clearly showing the utilization 共D + t1−兲共D + t2−兲共D + t3−兲共t兲
of fractional calculus. = E0D共D + 1−兲共D + 2−兲共t兲 共4.19兲
The complex modulus, denoted by Gⴱ共i ; 兲 for the third order
transition with one memory parameter is obtained by taking the The creep function for this second-order transition fluid, as de-
Fourier transformation of Eq. 共4.4兲, so that noted by J f 共t ; 兲, is
ⴱ
G 共i ; 兲/E0 = 兿
3
再 关共i兲 + n−兴
关共i兲 + tn 兴 − 冎兿
=
3
Gnⴱ共i ; 兲 再
J f 共t; 兲 = E0−1 1 +
t 1−t 2−t 3−
1−2−⌫共1 + 兲
t − J1共兲关1 − E关− 共t/1兲兴兴
冎
n=1 n=1
共4.14兲
− J2共兲关1 − E关− 共t/2兲兴兴 共4.20兲
where
关共i兲 + n−兴 where J1共兲 and J2共兲 are
Gnⴱ共i ; 兲 = = Gn⬘共兲 + iGn⬙共兲 共4.15兲
关共i兲 + tn−兴
J1共兲 = − 12共t1− − 1−兲共t2− − 1−兲共t3− − 1−兲/共1− − 2−兲
Thus, we obtain the real and imaginary parts of the complex
modulus for a third order transition as and
Downloaded 28 May 2010 to 137.104.114.22. Redistribution subject to ASME license or copyright; see http://www.asme.org/terms/Terms_Use.cfm
J2共兲 = − 22共t1− − 2−兲共t2− − 2−兲共t3− − 2−兲/共2− − 1−兲 3
冉 冊冋 冉 冊 冉 冊 册
monotonically increasing and looks like a rounded upward stair-
冉 冊
3
1/2
case starting at E0−1 , which is the initial response. The primary
兺
1 1 t t
J t; = E0−1 − E0−1 Jn 1 − exp erfc creep region occurs until time approaches 1 at which it will
2 n=1 2 n n
achieve a height of about E0−1J1共1兲. The staircase continues to
冋 冉冊 冉冊册
3
1/2 increase until it reaches a maximum height of J⬁. Also, for a fluid
兺J 2
1 t t
= J⬁ + E0−1 n exp erfc 共4.22兲 model, J⬁ will tend to infinity and E⬁ will tend to zero. The
n=1 n n mechanical model should represent the behavior of material with
memory since the memory parameter  stretches out these curves.
冉 冊 冉 冊冋 冉 冊 冉 冊 册
3
1/2 In summary, if the material is a solid, a second-order transition
兺
1 1 t t
G t; = E0 − E0 Gn 1 − exp erfc should be able to represent, within experimental error, the new
2 n=1 2 tn tn
data. The length of time of the experiment would have to be very
冉 冊冋 冉 冊 冉 冊 册
3
1/2 long to see the second transition. If the material is a fluid, then a
兺G
1 t t third order transition is need, but in its limiting case.
= E⬁ + E0 n exp erfc 共4.23兲
n=1 2 tn tn
where Jn共1 / 2兲 and Gn共1 / 2兲 are obtained from Eqs. 共4.10兲 and 5 Comments
共4.13兲, respectively. If the material does not age or flow, then these Rabotnov 共see pp. 201–209 in Ref. 关5兴兲 discussed aging using
equations should be adequate to reduce the experimental data. If nondifference kernels. For simplicity consider only the creep com-
the material flows, then Eqs. 共4.20兲 and 共4.17兲 are used to obtain
再
pliance when the specimen is in tension. The memory parameter
2t1−1/2t2−1/2t3−1/2 may be a function of the aging time, as denoted by ta if the
J f 共t;1/2兲 = E0−1 1 + t1/2 material ages. Thus, we may postulate that for aging the constitu-
冑 1−1/22−1/2 tive equation has the form
− J1共1/2兲关1 − exp共t/1兲erfc共t/1兲1/2兴
冎 冕
ta
共ta兲 J关共兲,ta − 兴d共兲 共5.1兲
+ − J2共1/2兲关1 − exp共t/2兲erfc共t/2兲1/2兴 共4.24兲 −⬁
3
n=1
Jk共t − 1,t − 2, . . . ,t − k兲 = ak 兿 J 共t − 兲
j=1
0 j 共5.2兲
Downloaded 28 May 2010 to 137.104.114.22. Redistribution subject to ASME license or copyright; see http://www.asme.org/terms/Terms_Use.cfm
Fig. 3 Decay curves
Fig. 2 Growth curves
M N
Downloaded 28 May 2010 to 137.104.114.22. Redistribution subject to ASME license or copyright; see http://www.asme.org/terms/Terms_Use.cfm
Acknowledgment Society, 46, pp. 651–656.
关24兴 Metzler, R., and Klafter, J., 2000, “The Random Walk’s Guide to Anomalous
Thanks to Patricia Hebert for her technical editorial help in Diffusion: A Fractional Dynamics Approach,” Phys. Rep., 339, pp. 1–77.
preparing this manuscript. Also, I would like to thank Professor 关25兴 Craiem, D., and Armentano, R., 2007, “A Fractional Derivative Model to
Describe Arterial Viscoelasticity,” Biorheology, 44, pp. 251–263.
Francesco Mainardi for helping me in clarifying my concepts of 关26兴 Nutting, P. G., 1921, “A New General Law of Deformation,” J. Franklin Inst.,
fractional calculus. 191, pp. 679–685.
关27兴 Catsiff, E., and Tobolsky, A. V., 1955, “Stress-Relaxation of Polyisobutylene
in the Transition Region,” J. Colloid Interface Sci., 10, pp. 375–392.
References 关28兴 Rabotnov, Y. N., 1953, “Some Problems of the Theory of Creep,” National
关1兴 Volterra, V., 1909, “Sulle Equazioni Integro-Differenziali della Theoria Advisory Committee for Aeronautics, Technical Memorandum No. 1353.
dell’Elasticità,” Atti Accad. Naz. Lincei, Cl. Sci. Fis., Mat. Natur., Rend., 18, 关29兴 Flügge, W., 1975, Viscoelasticity, 2nd ed., Springer-Verlag, Berlin
pp. 295–300. 关30兴 Zener, C. M., 1948, Elasticity and Unelasticity of Metals, University of Chi-
关2兴 Volterra, V., 1913, Lecons sur les Functions de Lignes, Gauthier-Villard, Paris. cago Press, Chicago.
关3兴 Riesz, F., and Sz.-Nagy, B., 1955, Functional Analysis, Frederick Ungar Pub- 关31兴 Tobolsky, A. V., and Catsiff, E., 1956, “Elastoviscous Properties of Polyisobu-
lishing, New York. tylene 共and Other Amorphous Polymers兲 From Stress-Relaxation Studies. IX.
关4兴 Gurtin, M. E., and Sternberg, E., 1962, “On the Linear Theory of Viscoelas- A Summary of Results,” J. Polym. Sci., 19, pp. 111–121.
ticity,” Arch. Ration. Mech. Anal., 11, pp. 291–356. 关32兴 Lee, G., Madigosky, W. M. and Eynck, J. J., 1979, “Dynamic Viscoelastic
关5兴 Rabotnov, Y. N., 1980, Elements of Hereditary Solid Mechanics, Mir Publish- Properties of Materials,” Naval Surface Weapons Center, Silver Springs, MD,
ers, Moscow. Report No. TR-78-138.
关6兴 Koeller, R. C., 1984, “Application of Fractional Calculus to the Theory of 关33兴 Caputo, M., and Mainardi, F., 1971, “A New Dissipation Model Based on
Viscoelasticity,” ASME J. Appl. Mech., 51, pp. 299–307. Memory Mechanism,” Pure Appl. Geophys., 91, pp. 134–147.
关7兴 Meshkov, S. I., Pachevskaya, G. N., Postnikov, V. S., and Rossikhin, U. A., 关34兴 Bagley, R. L., and Torvik, P. J., 1984, “On the Appearance of the Fractional
1971, “Integral Representations of Y-Functions and Their Application to Derivative in the Behavior of Real Materials,” ASME J. Appl. Mech., 51, pp.
Problems in Linear Viscoelasticity,” Int. J. Eng. Sci., 9, pp. 387–398. 294–298.
关8兴 Rossikhin, Y. A., and Shitikova, M. V., 2007, “Comparative Analysis of Vis-
关35兴 Bagley, R. L., and Torvik, P. J., 1983, “A Theoretical Basis for the Application
coelastic Models Involving Fractional Derivatives of Different Orders,” Frac-
of Fractional Calculus to Viscoelasticity,” J. Rheol., 27, pp. 201–210.
tional Calculus Appl. Anal., 10, pp. 111–121.
关36兴 Rogers, L., 1983, “Operators and Fractional Derivatives for Viscoelastic Con-
关9兴 Mainardi, F., and Gorenflo, R., 2007, “Time-Fractional Derivatives in Relax-
stitutive Equations,” J. Rheol., 27, pp. 351–372.
ation Processes: A Tutorial Survey,” Fractional Calculus Appl. Anal., 10, pp.
关37兴 Scott Blair, G. W., 1947, “The Role of Psychophysics in Rheology,” J. Colloid
269–308.
关10兴 Fraggstedt, M., 2006, “Power Dissipation in Car Tyres,” Thesis, Royal Insti- Sci., 2, pp. 21–32.
tute of Technology, Stockholm, Sweden. 关38兴 Koeller, R. C., 2007, “Toward an Equation of State for Solid Materials With
关11兴 Heibig, A., and Palade, L. I., 2008, “On the Rest State Stability of an Objective Memory by Use of the Half-Order Derivative,” Acta Mech., 191, pp. 125–
Fractional Derivative Viscoelastic Fluid Model,” J. Math. Phys., 49, p. 133.
043101. 关39兴 Heymans, N., and Podlubny, I., 2006, “Physical Interpretation of Initial Con-
关12兴 Beris, A. N., and Edwards, B. J., 1993, “On the Admissibility Criteria for ditions for Fractional Differential Equations With Riemann-Liouville Frac-
Linear Viscoelastic Kernels,” Rheol. Acta, 32, pp. 505–510. tional Derivatives,” Rheol. Acta, 45, pp. 765–771.
关13兴 Heymans, N., 2003, “Constitutive Equations for Polymer Viscoelasticity De- 关40兴 Podlubny, I., 1999, “Fractional Differential Equations,” Mathematics in Sci-
rived From Hierarchical Models in Cases of Failure of Time-Temperature Su- ence and Engineering, Vol. 198, Academic, New York.
perposition,” Signal Process., 83, pp. 2345–2357. 关41兴 Oldham, K. B., and Spanier, J., 1974, The Fractional Calculus, Academic,
关14兴 Read, B. E., 1989, “Mechanical Relaxation in Isotactic Polypropylene,” Poly- New York.
mer, 30, pp. 1439–1445. 关42兴 Ross, B., 1975, “A Brief History and Exposition of the Fundamental Theory of
关15兴 Sathiyanarayanan, S., Sivakumar, S., and Rao, C. L., 2005, “Experimental and the Fractional,” Lect. Notes Math., 457, pp. 1–36.
Modeling Study on Form 1 PVDF,” International Conference on Smart Mate- 关43兴 Miller, K. S., and Ross, B., 1993, An Introduction to the Fractional Calculus
rials Structures and Systems, Bangalore, India, Paper No. ISSS-2005/SA-26, and Fractional Differential Equations, Wiley, New York.
pp. 195–202. 关44兴 Gel’fand, I. M., and Shilov, G. E., 1964, Generalized Functions: Properties
关16兴 Dîkmen, Ü., 2004, “Modeling of Seismic Wave Attenuation in Soils by using and Operations, Vol. 1, Academic, New York.
Fractional Derivative Approach,” Ph.D. thesis, Department of Geophysical En- 关45兴 Koeller, R. C., 1986, “Polynomial Operators, Stieltjes Convolution, and Frac-
gineering, Ankara University, Turkey. tional Calculus in Hereditary Mechanics,” Acta Mech., 58, pp. 251–264.
关17兴 Agrawal, O. P., 1999, “An Analytical Scheme for Stochastic Dynamic Systems 关46兴 Caputo, M., and Mainardi, F., 1971, “Linear Models of Dissipation in Anelas-
Containing Fractional Derivatives,” ASME Paper No. DETC99/VIB-8238. tic Solids,” Riv. Nuovo Cimento, 1 pp. 161–198.
关18兴 Rossikhin, Y. A., and Shitikova, M. V., 2008, “Free Damped Vibration of a 关47兴 Mittag-Leffler, G. M., 1905, “Sur la représentation analytique d’une branche
Oscillator Based on Rabotnov’s Model,” Mech. Time-Depend. Mater., 12, pp. uniforme d’une function monogène,” Acta Math., 29, pp. 101–182.
129–149. 关48兴 Truesdell, C., and Noll, W., 1965, The Non-Linear Field Theories of Mechan-
关19兴 Mainardi, F., 1995, “The Time Fractional Diffusion-Wave Equation, ” Radio- ics, Encyclopedia of Physics, Vol. 3, Springer-Verlag, Berlin.
phys. Quantum Electron., 38, pp. 20–36. 关49兴 Erdélyi, A., Magnus, W., Oberhettinger, F., and Tricomi, F. G., 1957, Higher
关20兴 Mainardi, F., 1996, “The Fundamental Solution for the Fractional Diffusion- Transcendental Function, Vol. 3, McGraw-Hill, New York.
Wave Equation,” Appl. Math. Lett., 9, pp. 23–28. 关50兴 Welch, S. W. J., Rorrer, R. A. L., and Duren, R. G., 1999, “Application of
关21兴 Li, C., Liao, X., and Yu, J., 2003, “Synchronization of Fractional Order Cha- Time-Based Fractional Calculus Methods to Viscoelastic Creep and Stress Re-
otic Systems,” Phys. Rev. E, 68, p. 067203. laxation of Materials,” Mech. Time-Depend. Mater., 3, pp. 279–303.
关22兴 Maione, G., 2006, “A Digital, Noninteger Order, Differentiator Using Laguerre 关51兴 Bagley, R. L., 1991, “The Thermorheologically Complex Material,” Int. J.
Orthogonal Sequences,” Int. J. Intell. Contr. Syst., 11, pp. 77–81. Eng. Sci., 29, pp. 797–806.
关23兴 Hara, H., and Ikeda, N., 2005, “Nonlinear Fokker-Planck Equations on a 关52兴 Rabotnov, Y. N., 1948, “Some Problems of Creep Theory,” Vestn. Mosk.
Curved Spacetime Surface and Their Applications,” Journal of the Korean Univ., Ser. 1: Mat., Mekh., 10, pp. 81–91.
Downloaded 28 May 2010 to 137.104.114.22. Redistribution subject to ASME license or copyright; see http://www.asme.org/terms/Terms_Use.cfm