Download as pdf or txt
Download as pdf or txt
You are on page 1of 12

International Applied Mechanics, Vol. 41, No.

5, 2005

A METHOD FOR DETERMINING THE VISCOELASTIC CHARACTERISTICS


OF COMPOSITES

A. A. Kaminskii and M. F. Selivanov UDC 539.374,511.41

An efficient method for determining the deformation function of a composite is discussed. The method is
based on a fractional exponential representation of the deformation functions of the composite
components. The viscoelastic solution is obtained using the Volterra principle. The deformation function
is represented as a function of a base operator. Thus, the problem is solved by approximating the
deformation function by a continued fraction and applying the method of operator continued fractions.
A computational procedure is detailed and illustrated using data on longitudinal relaxation of
polymethylmethacrylate. As an example, the deformation of a polymethylmethacrylate-based fibrous
composite with viscoelastic properties is analyzed

Keywords: composite material, deformation function, operator function, operator continued fraction, Prony
series, viscoelasticity, viscoelastic characteristics

Introduction. Extensive use of polymers and polymer-based composites entails intensive development of the theory of
viscoelasticity needed to solve practical deformation problems for composites and boundary-value problems for materials with
time-dependent stress–strain state. Typical properties of these materials are relaxation (decreasing stresses at constant strains)
and creep (increasing strains at constant stresses). To describe the deformation of viscoelastic materials, Boltzmann proposed a
theory of hereditary viscoelasticity based on the superposition principle. According to this theory, the equation relating the
strains and stresses in a linear nonaging viscoelastic material under uniaxial loading and isothermal conditions reads
t
dσ(τ )
ε ( t ) = ∫ D( t − τ ) dτ, (1)

0

where ε(t) and σ(t) are the stresses and strains at time t, E is the elastic modulus, and D(ϑ) is the creep function. Integrating the
right-hand side of (1) by parts, we obtain the Volterra integral equation of the second kind describing creep in linear viscoelastic
media,
t
σ( t ) 1
E E∫
ε (t ) = + R ( t − τ ) σ(τ ) dτ, (2)
0

where R(ϑ) is the creep rate, which is sometimes called influence function, material function, or viscoelastic deformation
function; D ( 0) = 1/ E, and D′ ( ϑ ) = R ( ϑ ) / E.
Equation (2) may be written in operator form:

ε ( t ) = 1/ E (1+ R * ) ⋅ σ( t ), (3)

S. P. Timoshenko Institute of Mechanics, National Academy of Sciences of Ukraine, Kiev. Translated from Prikladnaya
Mekhanika, Vol. 41, No. 5, pp. 9–21, May 2005. Original article submitted September 2, 2004.

1063-7095/05/4105-0469 ©2005 Springer Science+Business Media, Inc. 469


where 1 + R* is the Volterra operator of the second kind described by the right-hand side of Eq. (2).
The physical characteristics of bodies possessing the property of creep are described by parameters that are invariant to
any processes of loading in a stationary temperature field. In the hereditary theory, such parameters are the elastic constants and
the parameters of the deformation function. The choice of the deformation function is an important step toward the viscoelastic
solution. The kernel of the operator (3) is chosen to be an analytical function containing parameters determined from
experimental data. To approximate experimental data, the following kernels are mainly used in practice:
(i) Duffing’s power kernel R( ϑ ) = λ / ϑ 1−β (0 ≤ β < 1);
(ii) Rzhanitsyn’s exponential power kernel R( ϑ ) = λ exp ( µ ϑ ) / ϑ 1−β (µ < 0, 0 ≤ β < 1); and
(iii) Rabotnov’s fractional exponential kernel R( ϑ ) = λϑ −α ∑ ∞n= 0 µ n ϑ n (1−α) / Γ [( n + 1)(1− α )] (µ < 0, 0 ≤ α < 1; Γ is
Euler’s gamma-function). A special case of this kernel for µ = –1 is Abel’s power kernel R( ϑ ) = λϑ −α / Γ (1− α ), which coincides
with Duffing’s kernel and for α = 0 with the exponential kernel R( ϑ ) = λ exp( µ ϑ ).
Each of the kernels may be used within the framework of a specific viscoelastic theory. Note that viscoelastic
deformation is sometimes described by sums of operators with kernels of one class. For example, Park and Schapery [17] used a
series of operators with exponential kernel (Prony series) as a deformation function. Such a choice of the creep function allows
us to get rid, without practically significant loss of accuracy, of the exponential kernel’s major shortcoming not typical of the
other kernels—inaccurate description of deformation at the initial stage. In [2, 4, 5], the viscoelastic deformation of composites
was described by a sum of operators with Rabotnov kernel. An important advantage of the exponential and fractional exponential
kernels is that operators with such kernels have a resolvent in the same class as the initial operator, which is also true of sums of
operators with such kernels. This allows the multistage inversion of operators, a process that underlies the method of operator
continued fractions. Thus, if the influence functions of the composite components are represented by sum of operators with
exponential or fractional exponential kernel, then we can obtain the deformation functions of the composite using an exact
analytical method.
1. Determining the Deformation Function of a Composite from the Deformation Functions of Its Components.
Let the composite components possess viscoelastic properties described by

⎛ ni ( i ) ⎞
E i* = E 0 ⎜⎜1+ ∑ λ j R * µ j
(i )
( )⎟⎟ ,
(i )
i = 1, 2, K , m, (4)
⎝ j= 1 ⎠

where R*(µ) is the Volterra integral operator of the first kind.


Note that we do not restrict the number of components with viscoelastic properties and the type of anisotropy. For
example, the deformation of the reinforcement or one of the materials of the matrix can be described by two operators E *j
(viscoelastic analogs of Young’s modulus and Poisson’s ratio) if it is isotropic and by five operators E *j if it is orthotropic.
Let us introduce a parameter κ:

κ = c⋅ max
i=1, 2 ,..., m
(µ (ji ) ), (5)
j=1, 2 ,..., nm

where the constant c (0 < c < 1) is selected close to unity in order that

κ> max
i=1, 2 ,..., m
(µ (ji ) ).
j=1, 2 ,..., nm

Using the algebra of resolvent operators [2], we can pass from the operator R*(µ) with the parameter µ to the following
operator with the parameter κ:

⎛ 1 ⎞ λ
λ R * ( µ ) = u ⎜1− ⎟, v = κ − µ, u= , (6)
⎝ 1+ vR (κ ) ⎠
* v

470
where κ < 0, u > 0, and v > 0, according to (5).
The norm of the operators with exponential or fractional exponential kernel is

| | R * (κ )| |= −1/ κ .

Consider an operator with a unit norm:

K * = −κ R * (κ ). (7)

Then

⎛ 1 ⎞ v
λR * = u j ⎜1− ⎟, w=− , (8)
⎝ 1+ wK * ⎠ κ

and the operator E* takes the form

⎛ n ⎛ 1 ⎞⎞ λ κ− µ j
= E 0 ⎜⎜1+ ∑ u j ⎜⎜1−
j
E* ⎟⎟ , uj = wj =
* ⎟⎟
, . (9)
⎝ j=1 ⎝ 1+ w j K ⎠ ⎠ κ− µ j κ

The reduced deformation characteristics of the composite as functions of time follow from the corresponding elastic
solution based on the Volterra principle and the method of operator continued fractions.
Let us determine the reduced mechanical characteristics of an elastic composite from the elastic characteristics of its
components using one of the numerous averaging method. Let a reduced elastic characteristic be

Ω = Ω( E j ), (10)

where Ej are the elastic characteristics of the composite components.


Then the time-dependent reduced deformation characteristics of the composite follow from the elastic solution (10)
based on the Volterra principle [4]:

( ) (
Ω * = Ω E *j = Ω E *j (K * ) . ) (11)

Thus, due to representation (9), the deformation functions of a composite modeled by a homogeneous viscoelastic
medium may be considered functions of the elastic characteristics of its components and one operator K * = –κR*(κ), which
enables us to apply the method of operator continued fractions.
2. Applying the Method of Operator Continued Fractions. The method is based on representing a real-valued
function Ω(x) corresponding to the operator function Ω (K*) as a continued fraction. The method is more efficient when the
function can be expanded into a proper continued fraction. Note that such a fraction can be derived from the Taylor-series
expansion of the function [6]. In this case, the convergents of the expansion possess a “fork” property: the exact value of the
function is between the values of two sequential convergents [5]. The approximating operator function obtained by the method
possesses the same property, which leads to mechanically correct results, even with a small order of approximation, when the
desired accuracy has not yet been reached. However, the complex form of the dependence Ω(x) in practical problems makes it
difficult to find the coefficients of its expansion into a proper continued fraction. In this case, the function Ω(x) may be subjected
to the Pade approximation Ωm(x) [1, 3], which coincides with Ω(x) at chosen points xk (k = 0, 1, …, m) of the interval of x. Such
an approach was used in [8, 9] in considering one initial fractional exponential deformation function (for a composite with elastic
reinforcement and viscoelastic isotropic matrix with elastic volume strain). In this case, high accuracy (about 0.001%) of
approximating the function Ω(x) was achieved with no more than five points xk, which resulted in a sum of two operators in the
approximating operator function Ω *m .
When the initial influence function is a sum of exponential or fractional exponential functions, the number m increases
because of a wide spread of µ in kernels (4). This leads to an increased order of the approximating fraction and, hence, to an
increased number of operators in Ω *m . The number of operators in Ω *m is minimized for prescribed accuracy by smoothing out the

471
maximum deviations from zero of the error function Ω (x) – Ωm(x), or the relative error function (Ω(x) – Ωm(x))/Ω(x) if the
spread in the values of Ω(x) is significant. The smoothing procedure is discussed below.
We will approximate a real-valued function corresponding to the operator function (10) by the continued fraction
m
x − x k −1 x − x0
Ω m ( x ) = V0 + D = V0 + , (12)
k =1 Vk x − x1
V1 +
V2 +
O
x − x m −1
+
Vm

where m is the number of subintervals of the interval of x; and x0, x1, …, xm are the points of division (x i ∈ [0, 1) since || K* || = 1).
Given values of the function Ω(x) at these points, the finite reciprocal differences are determined from the following recurrent
relations [1]:

V0 = Ω( x 0 ),

x k − x k −1
Vk = Vk ( x 0 , x1 ,K , x k−1 , x k ) = ,
Vk−1 ( x0 , x1 ,K , x k− 2 , x k ) −Vk−1 ( x0 , x1 ,K , x k− 2 , x k−1 )

where

V0 ( x ) = Ω( x ),

x − x k −1
Vk ( x 0 , x1 ,K , x k−1 , x ) = , k = 1, 2, …, m.
Vk−1 ( x0 , x1 ,K , x k− 2 , x ) −Vk−1 ( x0 , x1 ,K , x k− 2 , x k−1 )

According to the method of operator continued fractions, we will associate function (12) with the operator continued
fraction
m
* K * − xk
Ω m = 1+ D . (13)
k =1 Vk

Transforming it using the theorems of algebra of rezolvent operators [2], we obtain the approximation of the operator
function (11) in a convenient form:
[ m / 2]
Ω *m = 1+ ∑ (m )
λ j R* µ j ( (m )
), (14)
j= 1

where “[ ]” denotes the integer part.


3. Smoothed Deviations of the Error Function. Let us approximate a function f (x) defined on the interval [0, 1] by a
continued fraction F(x, x0, x1, …, xm) that coincides with f(x) at points x0, x1, …, xm [10] and has a positive range of values.
Mechanical considerations suggest that x0 = 0 (the value of the operator function describing the deformation of a material
coincides with the instantaneous value at initial time). We will select xs (s = 1, 2, …, m) so that the relative error of approximation

f ( x ) − F ( x, x s )
G ( x, x s ) = (15)
f (x )

has the smallest possible Chebyshev norm

||G ( x, x s )||= max | G ( x, x s )| , (16)


x∈ [ 0 ,1]

472
i.e., we will search for the minimum of (16) at the points xs. All involved functions are assumed continuous in xs and
differentiable in x.
Assume that the approximation problem has a uniquely defined solution X = x1, x2, …, xm, and G(x, xs) has the same
absolute values but alternating signs at m extreme points ξ0, ξ1, …, ξm of the interval [0, 1]. This solution can be obtained from
the equations

G (ξ 0 , x s ) + (−1) j+1 G (ξ j , x s ) = 0, j = 1, 2, …, m.

Let S be a set of vectors X for which the error G (x, xs) achieves different relative extrema with alternating signs at
(m + 1) points ξ0, ξ1, …, ξm. These extreme points are considered functions of xs : ξ j = ξ j(xs). At the internal extreme points,
these functions satisfy the equation

⏐∂G ( x, x s ) ⏐
⏐ ⏐ = 0.
⏐ ∂ x ⏐x=ξ
j

Substituting these functions into the expression

(
p j ( x s ) = G (ξ 0 ( x s ), x s ) + (−1) j+1 G ξ j ( x s ), x s ,) j = 1, 2, …, m, (17)

we obtain m nonlinear equations for xk, p j ( x s ) = 0, j = 1, 2, …, m. This system of equations can be solved by Newton’s method.
If pj (xs) are considered to be the coordinates of the vector TX, then relations (17) define an operator T. Beginning from
X , we sequentially determine the approximations X(n) by the formula
(0)

−1
X ( n+1) = X ( n ) − ⎡⎣T(′X (n) ) ⎤⎦ TX ( n ) , n = 0, 1, 2, … .

The Frechet derivative of the operator T is a matrix with the elements

1 ∂F (ξ 0 ( x s ), x s ) 1 ∂F (ξ j ( x s ), x s )
t jk = − − (−1) j+1 , j, k = 1, 2, …, m.
f (ξ 0 ) ∂x k f (ξ j ) ∂x k

It is convenient to represent the initial vector X (0) as

i
xi : f ( xi ) = f ( 0) + (1− ε )( f (1) − f ( 0)), i = 0, 1, …, m.
m

The correction vector h = (h1, h2, …, hm) follows from the system of linear equations −T(′X (0) ) h = TX ( 0 ) ,
m
T(′X (0) ) = ( t jk ) or −∑ t jk hk = p j ( X ( 0 ) ).
k =1
Thus, the subsequent approximation is X (1) = X (0) + h. Continuing the iterative process, we arrive at the solution X(n).
4. Numerical Examples. Let the deformation of a composite’s components be described using the Prony series.
4.1. Deformation of Polymethylmethacrylate. The relaxation characteristics Ee, Ej, and, ρj (j = 1, 2, …; n, n = 11) of
polymethylmethacrylate (PMMA) are listed in [17]. The relaxation modulus has the form
n
E ( t ) = E e + ∑ E j exp (−t / ρ j ).
j= 1

The relations

1 Ej
E0 = Ee + ∑ E j , µ j =− , λj = µj
ρj E0

473
TABLE 1 TABLE 2

j λ j , day–1 µ j , day–1 Set 1 Set 2


j
λj µj λj µj
1 –3.74⋅105 –4.32⋅106
1 3.60⋅105 –3.95⋅106 3.60⋅105 –3.95⋅106
2 –5.46⋅104 –4.32⋅105
2 6.09⋅104 –3.69⋅105 6.12⋅104 –3.74⋅105
3 –1.07⋅104 –4.32⋅104
3 1.51⋅104 –3.00⋅104 1.52⋅104 –3.02⋅104
4 –1.16⋅103 –4.32⋅103
4 3.26⋅103 –2.27⋅103 3.26⋅103 –2.27⋅103
5 –74.8 –4.32⋅102
5 7.47⋅102 –1.65⋅102 7.44⋅102 –1.68⋅102
6 –3.01 –43.2
6 2.01⋅102 –13.1 2.06⋅102 –13.2
7 –7.91⋅10–2 –4.32
7 70.2 –1.43 69.5 –1.44
8 –2.66⋅10–3 –4.32⋅10–1 8 21.3 –1.47⋅10–1 21.7 –1.47⋅10–1
9 –7.10⋅10–5 –4.32⋅10–2 9 6.20 –2.02⋅10–2 6.11 –2.03⋅10–2

10 –1.52⋅10–6 –4.32⋅10–3 10 5.21⋅10–1 –3.36⋅10–3 5.44⋅10–1 –3.39⋅10–3

11 –1.85⋅10–7 –4.32⋅10–4 11 9.50⋅10–2 –2.93⋅10–4 9.51⋅10–2 –2.98⋅10–4

allow conversion from the characteristics Ee, Ej, and ρj to those from (4):
t
⎛ n ⎞
E (t ) = E * ⋅ 1, E* = E 0 ⎜⎜1+ ∑ λ j R * ( µ j )⎟⎟ , R * (µ j ) ⋅ 1= ∫ exp ( µ j ϑ ) dϑ. (18)
⎝ j= 1 ⎠ 0

These characteristics, λj and µj (j = 1, 2, …, n), are summarized in Table 1 (E 0 = 2.24 ⋅ 106 MPa, 1 day = 8.64 ⋅ 104 sec,
and 1 MPa = 103 N/m2).
Park and Schapery [17] proposed a numerical method for interconversion between the relaxation characteristics and the
creep characteristics (glassy compliance Dg, retardation strengths Dj, and retardation times τj):
n
(
D ( t ) = Dg + ∑ D j 1− exp −t / τ ( j )).
j= 1

In terms of the algebra of resolvent operators, this interconversion consists in inversion of the operator E* in (18). In this
case, we have
1 1 1
D* = = , (19)
E* E0 n
1+ ∑ λ j R* (µ j )
j= 1

or, applying the sum inversion theorem for resolvent operators, we obtain

⎛ n (2 ) ⎞
D * = D0 ⎜⎜1+ ∑ λ j R * µ j
(2 )
( )⎟⎟ , (20)
⎝ j= 1 ⎠

474
D(t) lgD(t)

500

0
0 0.5 lg t, day –5 0 lg t, day

a b
Fig. 1

(2 )
where D0 = 1/E0, µ j (j = 1, 2, …, n) are the roots of the equation

n λj
∑µ −µ
= 1, (21)
j= 1 j

(2 )
and λ j (j = 1, 2, …, n) is the solution of the system of linear algebraic equations

(2 )
n λj
∑ (2 )
= 1, i = 1, 2,... , n.
j= 1 µ i −µ j

Table 2 summarizes the creep characteristics obtained by Schapery and by us (D0 = 4 .47⋅ 10−7 ). Since the accuracy of
the roots of Eq. (21) is much higher than that of the initial data, solution (20) may be regarded as exact. Note that the values of the
deformation functions D(t) = D*·1 calculated from set 1 (the solution from [7]) and set 2 (the solution based on the algebra of
resolvent operators) differ by less than 0.2%. Semilog and log-log plots of the function D(t) are shown in Fig. 1.
Let us apply the method described in Sect. 1. We will represent operator (20) as an operator function of a base resolvent
operator. According to (5), we select the parameter κ = – 2.9⋅10–4 and determine the coefficients uj and wj of (9). Their
coefficients are given in Table 3.
Thus, operator (20) has the form

⎛ n ⎛ 1 ⎞⎞
D * = D0 ⎜⎜1+ ∑ u j ⎜⎜1− ⎟⎟ ,
* ⎟⎟
(22)
⎝ j= 1 ⎝ 1+ w j K ⎠⎠

t
where K* ⋅ 1= −κR * (κ) ⋅ 1= −κ ∫ exp (κϑ ) dϑ.
0
If the operator function describing the deformation of a multicomponent composite is a function of several operators
(20), then the parameter κ < 0 is selected, according to (5), to be larger than all parameters of sets of characteristics µ for the
composite components. The deformation functions of the other components are represented in the form (22) with identical
operator K* and sets of characteristics uj and wj. The deformation function of the composite becomes a function of one operator
K*, which makes it possible to apply the method of operator continued fractions. According to this method: (i) the function of K*
is associated with a function of real variable, (ii) the function of real variable is approximated by a continued fraction, and (iii) the
real variable in the continued fraction is replaced by K* and the algebra of resolvent operators is applied. This yields a sum of
base operators that approximates the initial operator function. Note that this method is most efficient when the deformation
function is complex. If it is rational, then the algebra of resolvent operators can be applied directly.

475
TABLE 3 TABLE 4

j uj wj j xj Vj j xj Vj

1 9.11⋅10–2 1.36⋅1010
1 0 1 10 2.22⋅10–5 1.51⋅10–6
2 1.64⋅10–1 1.29⋅109
2 9.13⋅10–11 1.24⋅10–9 11 8.20⋅10–5 61.1
3 5.03⋅10–1 1.04⋅108
3 5.43⋅10–10 2.60⋅10–1 12 2.89⋅10–4 6.40⋅10–6
4 1.44 7.84⋅106

5 4.43 5.79⋅105 4 2.57⋅10–9 1.29⋅10–8 13 9.99⋅10–4 1.72⋅102

6 15.6 4.55⋅104 5 1.11⋅10–8 8.31⋅10–1 14 3.45⋅10–3 3.24⋅10–5


7 48.1 4.98⋅103
6 4.87⋅10–8 7.33⋅10–8 15 1.22⋅10–2 2.92⋅102
8 1.48⋅102 5.05⋅102
7 2.29⋅10–7 3.04 16 4.63⋅10–2 4.44⋅10–4
9 3.06⋅102 69.0
8 1.18⋅10–6 3.72⋅10–7 17 2.05⋅10–1 2.79⋅102
10 1.75⋅102 10.7

11 1.14⋅104 2.88⋅10–2 9 5.49⋅10–6 15.0 18 7.95⋅10–1 2.47⋅10–3

Let us demonstrate the capabilities of the method by way of an example. According to step (i), we associate function
(22) with the following function of real variable:

⎛ n ⎛ 1 ⎞ ⎞⎟
f ( x ) = D0 ⎜⎜1+ ∑ u j ⎜1− ⎟⎟ . (23)
⎝ j=1 ⎝ 1+ w j x ⎠ ⎠

As steps (ii) and (iii), we pass to (20) based on the known function f(x) and the known parameter κ of the operator K*.
Since f (x) is a rational function, it can be analytically represented by a continued fraction at step (ii). Note that we go through step
(ii) just to illustrate the method, since the unknown function is simultaneously initial one, which allows us to determine the
computational error. Now we apply the approximation procedure outlined in Sect. 3. Let the approximating function F (x, xs)
(s = 1, 2, …, m) be Ωm(x) (12). The partial derivatives of this function with respect to xs are determined numerically. The
parameters of the continued fraction Ωm(x) (12) with m = 18 are summarized in Table 4. Figure 2 shows plots of the function f(x)
and error function G(x, xs) (15). Note that f(x) and Ωm(x) coincide for m = 23. At step (iii), we determine the operator function
Ω *m = Ωm(K*) to obtain an approximation in the form (14). The coefficients of this approximation are collected in Table 5.
Figure 3 shows the error function G(t) = (D*·1 – Ω *m ·1)/D*·1. Thus, the errors of the function of real variable and the
corresponding function of time differ by an order of magnitude. This requires more accurate approximation of the function f(x).
To this end, we may increase the number of points of division xs.
4.2. Deformation of Composite. Consider a PMMA-matrix composite unidirectionally reinforced with organic
filaments. The viscoelastic deformation of the filaments is described by the following exponential function:
t
D* = D0 ( 1+ λ(1) R * ( µ (1) )), R* ( µ) ⋅ 1= ∫ exp ( µϑ )dϑ , D0 = 1/ E 0 .
0

The deformation parameters have the following values [12]: E = 140 GPa, λ = 1.19⋅10–2 day–1, and µ = –9.5⋅10–3 day–1.
According to (9), we obtain

476
f G ⋅103

500 0

–2

–4
0 0.5 x –10 –5 lg x

a b
Fig. 2

TABLE 5

j λj µj

1 3.41⋅105 –2.21⋅106

2 2.90⋅104 –6.29⋅104

3 4.41⋅103 –2.93⋅103

4 7.91⋅102 –1.43⋅102

5 1.88⋅102 –7.97

6 50.9 –6.46⋅10–1

7 14.4 –5.50⋅10–2

8 2.03 –6.84⋅10–3

9 9.81⋅10–2 –2.90⋅10–4

⎛ ⎛ 1 ⎞⎞
D * = D0 ⎜1+ u ⎜1− ⎟⎟ , (24)
⎝ ⎝ 1+ wK * ⎠ ⎠

λ 1.19⋅ 10−2 κ − µ 2.9⋅ 10−4 − 9.5⋅ 10−3


where u = = = 1.29 and w = = = 31.8.
κ− µ 2.9⋅ 10−4 + 9.5⋅ 10−3 κ 2.9⋅ 10−4
Thus, the functions (22) and (24) describe the deformation of the matrix and reinforcement, respectively.
Let the x3-axis be directed along the filaments. The materials of the filaments and matrix are assumed isotropic. The
volume densities of the reinforcement and matrix are c1 and c2 = 1 – c1, respectively.
The physicomechanical characteristics of a material reinforced with unidirectional, randomly laid filaments form a
statistically isotropic field in a plane perpendicular to the reinforcement direction. Assuming that the properties and parameters

477
G ⋅102

–1

–2
–5 0 lg t, day
Fig. 3

of the material are statistically homogeneous at distances far exceeding the typical length scale of inhomogeneities, we can
consider the composite as a homogeneous material with effective characteristics [13]. We also assume that the property of
ergodicity is valid, i.e., the volume (square) average is equal to the statistical average.
Since the properties of the material are invariant to rotations of the coordinate system around the x3-axis (symmetry
axis) by arbitrary angles, the material is transversely isotropic. The relationship between the mean stresses and strains in a
transversely isotropic composite is expressed as

σ11 = λ 11ε 11 + λ 12 ε 22 + λ 13 ε 33 , σ 23 = λ 44 γ 23 ,
σ 22 = λ 12 ε 11 + λ 11ε 22 + λ 13 ε 33 , σ13 = λ 44 γ 13 , (25)
σ 33 = λ 13 ε 11 + λ 13 ε 22 + λ 33 ε 33 , σ12 = λ 66 γ 12 ,

and λ66 = (λ11 – λ12)/2.


The effective constants are defined [13] by

1 c1 c 2 (λ 1 + µ1 − λ 2 − µ 2 ) 2
(λ 11 + λ 12 ) = 〈λ + µ〉 − ,
2 c1 (λ 2 + µ 2 ) + c 2 (λ 1 + µ1 ) + µ c

c1 c 2 (λ 1 − λ 2 )(λ 1 + µ1 − λ 2 − µ 2 )
λ 13 = 〈λ 〉 − ,
c1 (λ 2 + µ 2 ) + c 2 (λ 1 + µ1 )+ µ c

c1 c 2 (λ 1 − λ 2 ) 2
λ 33 = 〈λ + 2µ〉 − ,
c1 (λ 2 + µ 2 ) + c 2 (λ 1 + µ1 ) + µ c

c1 c 2 ( µ1 − µ 2 ) 2 c1 c 2 ( µ1 − µ 2 ) 2
λ 44 = 〈 µ〉 − , λ 66 = 〈 µ〉 − , (26)
c1 µ 2 + c 2 µ1 + µ c λc +µc
c1 µ 2 + c 2 µ1 + µ c
λ c + 3µ c

where the constants λc and µc in (26) determine how well the components are bound and how close the effective constants are to
the real values. For a fiber-reinforced composite with K1 > K2 and µ1 > µ2, we have

−1 −1
1 2 1 2
λc = − µc, µc = , K = Λ+ Μ, (27)
K 3 µ 3

where Λ and Μ are the Lamé constants,

478
lg (1/Eij)
1/G13
2.5
1/G12
2.0

1.5
1/E11
1.0

0.5

0
8 6 4 2 0 2 4 lg t, day

Fig. 4

Ei νi Ei
Λi = , Μi = , i = 1, 2. (28)
(1+ ν i )(1− 2ν i ) 2(1+ ν i )

where Ei and νi are the elastic characteristics of the isotropic material, Ei are Young’s moduli, νi are Poisson’s ratios, i = 1 refers
to the reinforcement and i = 2 to the matrix, and the averaging operator “〈 〉” is defined by 〈 X 〉 = c1X1 + c2X2 (X is an elastic
constant).
Resolving relations (25) for the macrostrains, we get

1 ν12 ν13 1
ε 11 = σ11 − σ 22 − σ , γ 23 = σ ,
E11 E11 E 33 33 G13 23
ν12 1 ν13 1
ε 22 = − σ11 + σ 22 − σ , γ 13 = σ , (29)
E11 E11 E 33 33 G13 13
ν13 ν13 1 1
ε 33 = − σ11 − σ + σ , γ 12 = σ .
E11 E11 22 E 33 33 G12 12

The technical constants of the composite E11, E33, ν12, ν13, G12, and G13 are expressed in terms of the characteristics
(26) as

∆ ∆ ∆ 12 ∆ 13
E11 = , E 33 = , ν12 = , ν13 = , G12 = λ 66 , G13 = λ 44 , (30)
∆ 11 ∆ 33 ∆ 11 ∆ 33

where ∆ 11= λ 11λ 33 − (λ 13 ) 2 , ∆ 12 = λ 12 λ 33 − (λ 13 ) 2 , ∆ 13 = λ 13 (λ 11 − λ 12 ), ∆ 33 = (λ 11 ) 2 − (λ 12 ) 2 , and ∆ = λ 33 ∆ 33 − 2λ 13 ∆ 13 .


To solve the linear viscoelastic problem in the quasistatic case, we will use the Volterra principle [4]. It allows us to
reduce the problem to the determination of an operator function. This function is obtained from the corresponding elastic
problem by substituting viscoelasticity operators for the corresponding elastic characteristics.
Since the materials of the reinforcement and matrix are isotropic, the deformation of each component is described by
two viscoelasticity operators, E i* and ν *i . If the volume deformation of the components is elastic, then

1− 2ν * 1− 2ν
*
= .
E E

The deformation characteristics become

479
1 1 ⎛ E* ⎞
E* = , ν* = ⎜1− (1− 2ν)⎟ , (31)
D* 2⎝ E ⎠

where the function D* is defined by (22) for the matrix and by (20) for the reinforcement. Thus, the operators (31) are viscoelastic
analogs of Young’s moduli and Poisson’s ratios of the components in (28). Replacing the elastic characteristics in (28), (27),
(26), and (30) by their viscoelastic analogs, we obtain the functions of E11 * , E * , ν * , ν * , G * , and G * as functions of the
33 12 13 12 13
operator K*. Determining these functions using the procedure illustrated in Sect. 4.1, we obtain their representations as sums of
base operators (Prony series). Functions describing some time-dependent deformation characteristics are shown in Fig. 4.

REFERENCES

1. P. I. Bodnarchuk and V. Ya. Skorobagat’ko, Branching Continued Fractions and Their Application [in Russian],
Naukova Dumka, Kiev (1974).
2. A. N. Guz, A. A. Kaminskii, V. M. Nazarenko, et al., Fracture Mechanics, Vol. 5 of the 12-volume series Mechanics of
Composite Materials [in Russian], A.S.K., Kiev (1996).
3. W. B. Jones and W. J. Thron, Continued Fractions, Addison-Wesley Publ., Massachusetts (1980).
4. A. A. Kaminskii, Fracture of Viscoelastic Bodies with Cracks [in Russian], Naukova Dumka, Kiev (1990).
5. A. A. Kaminskii and S. A. Kekukh, “Method of solving problems of the linear theory of viscoelasticity for anisotropic
materials (in the presence of cracks),” Int. Appl. Mech., 30, No. 4, 320–327 (1994).
6. A. A. Kaminskii and M. F. Selivanov, “Long-term fracture of a laminated composite with a crack under time-dependent
loading,” Mekh. Komp. Mater., 36, No. 4, 545–558 (2000).
7. A. A. Kaminskii and M. F. Selivanov, “Deformation of a viscoelastic composite plate with a crack,” Visn. Donetsk.
Univ., No. 2, 42–45 (2002).
8. A. A. Kaminskii and M. F. Selivanov, “Viscoelastic deformation of a reinforced plate with a crack,” Int. Appl. Mech., 38,
No. 12, 1508–1517 (2002).
9. A. A. Kaminskii and M. F. Selivanov, “Creep-induced stress redistribution around an elliptic opening in a viscoelastic
orthotropic plate,” Dop. NAN Ukrainy, No. 6, 49–54 (2004).
10. Von L. Collatz, Functional Analysis and Numerical Mathematics, Acad. Press, New York (1966).
11. Yu. N. Rabotnov, Elements of the Hereditary Mechanics of Solids [in Russian], Nauka, Moscow (1977).
12. A. M. Skudra and F. Ya. Bulavs, Strength of Reinforced Plastics [in Russian], Khimiya, Moscow (1982).
13. L. P. Khoroshun, B. P. Maslov, E. N. Shikula, and L. V. Nazarenko, Statistical Mechanics and Effective Properties of
Materials, Vol. 3 of the 12-volume series Mechanics of Composite Materials [in Russian], Naukova Dumka, Kiev
(1993).
14. G. V. Gavrilov, “Subcritical growth of an internal circular crack in an aging viscoelastic laminated composite,” Int. Appl.
Mech., 40, No. 1, 77–82 (2004).
15. A. A. Kaminskii and M. F. Selivanov, “A method for solving boundary-value problems of linear viscoelasticity for
anisotropic composites,” Int. Appl. Mech., 39, No. 11, 1294–1304 (2003).
16. A. A. Kaminskii and M. F. Selivanov, “Influence of cyclic load on crack growth kinetics in a viscoelastic orthotropic
plate made of a composite material,” Int. Appl. Mech., 40, No. 9, 1037–1041 (2004).
17. S. W. Park and R. A. Schapery, “Methods of interconvention between linear viscoelastic material functions. Part I—A
numerical method based on Prony series,” Int. J. Solids Struct., 36, 1653–1675 (1999).

480

You might also like