A boundary element methodology for viscoelastic analysis:

Part I with cells

A.D. Mesquita, H.B. Coda
*
Department of Structural Engineering, Sao Carlos School of Engineering, Sao Paulo University,
Av. Trabalhador Sao Carlense 400, 13566-960 Sao Paulo, Sao Carlos, Brazil
Received 1 March 2003; received in revised form 1 December 2005; accepted 12 April 2006
Available online 7 July 2006
Abstract
In this study Kelvin and Boltzmann viscoelastic models are implemented in a two-dimensional boundary element atmo-
sphere. This general methodology is based on dierential constitutive relations for viscoelasticity, avoiding the use of relax-
ation functions. Part I describes a methodology using internal cells. This methodology makes it possible to consider viscous
parameters, which are not proportional to elastic tensor. From the kinematical relation between material and strain velo-
cities at the approximation level a simple time marching process is achieved. At the end of Part I, numerical examples
are provided to validate the methodology. The BEM viscoelastic formulation without using cells is carefully described
in Part II.
2006 Elsevier Inc. All rights reserved.
1. Introduction
One can nd in the literature a great amount of computational techniques to solve various similar mechan-
ical problems. It is not dierent when viscoelasticity is the subject of study. A brief description of some well-
known procedures is given below.
The rst classic procedure for solving viscoelastic problems is the use of mathematical transformations. The
viscoelastic equation can be transformed into pseudo-elastic one by means of LaplaceCarson transforms.
After solving the transformed problem, a numerical inversion can be performed recovering the desired time
domain behaviour [14]. This procedure presents some diculties when viscous parameters vary along time,
or when complicated time dependent boundary conditions are imposed.
Incremental schemes are also available in the literature. This kind of procedure uses relaxation functions to
transform the convolutional aspect of the viscous behaviour in discrete contributions added to the elastic
response [59]. In some studies viscoelastic problems are solved as viscoplastic ones, i.e., introducing the
0307-904X/$ - see front matter 2006 Elsevier Inc. All rights reserved.
doi:10.1016/j.apm.2006.04.006
*
Corresponding author. Tel.: +55 16 33739482; fax: +55 16 2739482.
E-mail addresses: [email protected] (A.D. Mesquita), [email protected] (H.B. Coda).
Applied Mathematical Modelling 31 (2007) 11491170
www.elsevier.com/locate/apm
viscous behaviour by means of residuals calculated at the eective stress level [1013]. The complete informa-
tion about creep (and relaxation) approaches can be seen in Refs. [14,15].
Recently, the authors developed an alternative FEM methodology to analyse bodies following the Kelvin
and Boltzmann models [16,18]. In this study this new methodology is extended to BEM formulation. The
development of such subject follows two dierent ways. The rst, described in Part I, follows exactly the same
idea of the FEM approach (respecting the particularities of BEM), i.e., the kinematical relation between the
material velocity and the strains velocity at the approximation level is used resulting in BEM cells. In Part II a
more elaborated procedure is described avoiding the use of cells for both Kelvin and Boltzmann models. A
simple coupling between FEM and BEM is also described in Part II to make it possible to analyse reinforced
media, as for example tunnels and sandwich panels.
The division of the subject into two parts is necessary as the rst shows the basis of the methodology and
gives a background to the proposed strategy, while the second, avoiding the use of cells, makes it possible to
analyse innite and half-space regions.
It is important to observe that both formulations are able to consider the material with time varying viscous
properties, the so-called ageing viscoelasticity. At the end of each part examples are provided in order to val-
idate the technique and check its accuracy and stability.
2. Denition of the problem
Before describing the numerical procedure to analyse the mechanical behaviour of solids presenting visco-
elastic characteristics it is necessary to dene the related dierential equilibrium equation and boundary con-
ditions. A general solid X subjected to prescribed surface forces p and displacement restrictions u can be seen
in Fig. 1.
From Fig. 1 one observes that the boundary of the solid is divided into two parts C
1
and C
2
. Over C
1
and C
2
one prescribes surface forces and displacement restrictions, respectively. This can be written in a more formal
way dening the boundary conditions of the body as follows:
u
i
x; t u
i
x; t over C
1
1
p
i
x; t p
i
x; t over C
2
2
where x is a point over the boundary and t is time. The desired solutions of the problem are the displacement
eld u(x, t) "x 2 X and the stress eld r(x, t) "x 2 X.
The usual way to write the dierential equation that governs the solid is the study of the equilibrium of an
innitesimal part of it, see Fig. 2.
From Fig. 2, following Einstein notation, one writes:
r
ij;i
b
j
qu
j
c_ u
j
3
1
2
1
2
n
Fig. 1. Problem description.
1150 A.D. Mesquita, H.B. Coda / Applied Mathematical Modelling 31 (2007) 11491170
where _ u
i
is the velocity, u
i
is the acceleration, c is the friction parameter (see Section 2), q is the mass density
and b
i
is the body force. One should use the Cauchy formulae to relate directly stress to surface force, as,
p
i
r
ij
n
j
4
where n
j
is the surface outward unit vector.
To complete the description of the problem one applies the kinematical hypothesis that relates displacement
and velocity to strain and strain rate. After that one applies the constitutive equation that relates strain and
strain rate to stress, transforming Eq. (3) as a function of displacement and its derivatives only. These relations
are shown in the following sections, but the numerical strategy to solve the physical problem is based on the
equilibrium equation written for stress, i.e., Eq. (3).
3. Dierence among viscosity, friction and inertia
In order to develop a methodology to solve viscoelastic problems based on dierential constitutive relations
for viscoelasticity (carrying out the time integration after introducing the space approximations), one has to
understand the equilibrium relations for an innitesimal portion of the studied body, briey introduced in the
achievement of the dierential equation in the previous section. In Fig. 3ac dashed arrows represent the
required forces to support the eects indicated by solid arrows. Therefore, taking into account Fig. 3a, one
can realise that the force qu
i
dV is required to impose acceleration u
i
to an innitesimal part of the body
dV with mass density q. If the internal friction c is assumed independent of elastic relations, it is necessary
to impose a force c_ u
i
dV to support a movement with constant velocity _ u
i
(see Fig. 3).
Assuming the viscous eect g also independent of the elastic relations, opposite forces equal to g_ e
i
dA (sim-
plied notation) are necessary to support the relative velocity between neighbouring points, Fig. 3c. Inertial
and friction forces are volume-type forces, while elastic and viscoelastic forces are stress-type ones.
dx
2
1
x d
12 +
12
x
1
dx
1
11 + d
1
x
11
1
x
21 + d
2
x
21
2
x
22
d + 2
x
22
2
x
11
12
22
21
b
2
b
1
c
1
1
2
1
c
u
u
u
u
Fig. 2. Equilibrium of an innitesimal part of the solid.
u
udv
v d
dv
u
v d u c
u
u
+
u
x
dx
dA A d
d
u
x
A
d
u
x
A
(a) Acceleration (b) Friction (c) Viscosity
Fig. 3. One-dimensional representation of a moving point.
A.D. Mesquita, H.B. Coda / Applied Mathematical Modelling 31 (2007) 11491170 1151
Considering continuity properties, the viscous stresses appearing along the right side of an innitesimal ele-
ment should be generated by the neighbouring innitesimal element placed on its right side. Along the bound-
ary this compatibility is naturally assured by boundary conditions, Eqs. (1) and (2) and Cauchy formulae (4).
At this point, viscous eects should be incorporated into the global equilibrium equation of the body taking
into account strain velocity. Moreover, the viscous characteristics of the body must satisfy boundary condi-
tions together with the elastic ones. In order to fulll these requirements the following sections describe the
introduction of viscous stress into Eq. (3).
4. Basic relations for viscoelasticity
This section is divided into two main parts, one related to the Kelvin model and the other related to the
Boltzmann standard relations.
4.1. Kelvin model
Using rheological models dened in the uniaxial space is the usual way adopted to describe the viscoelastic
behaviour of solids. A simple representation, usually adopted to describe this kind of behaviour is the Kelvin
Voigt viscoelastic model, Fig. 4.
From this model the following relations are stated:
e
ij
e
e
ij
e
v
ij
5
r
ij
r
e
ij
r
v
ij
6
where e
ij
and r
ij
are the strain and stress tensors. Superscripts v and e represent viscous and elastic parts,
respectively.
The elastic stress can be written in terms of strain components, as follows:
r
e
ij
C
ijlm
e
e
lm
C
ijlm
e
lm
7
Similarly, the following relation gives the viscous stress components, see Fig. 3c,
r
v
ij
g
ijlm
_ e
v
lm
g
ijlm
_ e
lm
8
In Eqs. (7) and (8), C
ijlm
and g
ijlm
are the elastic tensor and the viscous constitutive parameters, respectively,
dened as follows [17]:
C
ijlm
kd
ij
d
lm
ld
il
d
jm
d
im
d
jl
9
g
ijlm
h
k
kd
ij
d
lm
h
l
ld
il
d
jm
d
im
d
jl
10
where k and l are the Lame`s constants, given by:
k
mE
1 m1 2m
11
l G
E
21 m
12
E

Fig. 4. KelvinVoigt viscoelastic model (uniaxial representation).

1152 A.D. Mesquita, H.B. Coda / Applied Mathematical Modelling 31 (2007) 11491170
in which E and m are the Young modulus and Poisson ratio, respectively, while h
k
and h
l
are the viscosity coef-
cients. It is worth noting that these coecients can vary along time, as described in Appendix A, and that the
formulation presented here is valid for any time variation. As h
k
is not necessarily equal to h
l
, the viscous
parameter tensor is not proportional to the elastic compliance tensor.
Replacing Eqs. (7) and (8) into Eq. (6) gives
r
ij
C
ijlm
e
lm
g
ijlm
_ e
lm
13
This is the general Kelvin constitutive relation used to generate the BEM formulation proposed here.
Substituting Eq. (6) into Eq. (3) and neglecting the dynamic terms, one has
r
e
ij;i
r
v
ij;i
b
j
0 14
or simply
r
ij;i
b
j
0 15
Note that Eq. (14) explicitly exhibits the viscous stress term which plays an important role in the studied
problem.
4.2. Boltzmann model
Another representation employed to describe the mechanical behaviour of viscoelastic materials, stress/
strain constitutive relation, is the so-called standard Boltzmann model. This model is more general than
the previous one and can be described in a uniaxial representation as illustrated in Fig. 5.
This model is represented by a serial arrangement of a KelvinVoigt model and an elastic relation. It can
reproduce both instantaneous and viscous behaviours of a specic material.
It is easy to observe (see Fig. 5) that the stress level for each part of the model, elastic and viscoelastic, is the
same:
r
ij
r
e
ij
r
ve
ij
16
where r
ij
; r
e
ij
and r
ve
ij
are, respectively, total, elastic and viscoelastic stress parts. The total strain can be
decomposed into its elastic and viscoelastic parts, i.e.:
e
lm
e
e
lm
e
ve
lm
17
From Fig. 5, one can observe that the viscoelastic stress is the summation of a viscous and an elastic part, as
follows:
r
ve
ij
r
el
ij
r
v
ij
18
where r
v
ij
is the viscous part and r
el
ij
is the elastic part of the stress developed in the KelvinVoigt fragment of
the Boltzmann model.
From the previous equations, one is able to dene the dierential viscoelastic constitutive relation used here
to build the desired boundary integral equations.
r
qs

E
ve
E
e
E
ve
E
e
_ _
~
C
qsck
e
ck

E
e
E
ve
E
ve
E
e
_ _
~g
qsck
_ e
ck

E
ve
E
ve
E
e
_ _
~g
qsij
~
D
ijck
_ r
ck
19
E
e
E
ve

e ve
Fig. 5. Boltzmann viscoelastic model (uniaxial representation).
A.D. Mesquita, H.B. Coda / Applied Mathematical Modelling 31 (2007) 11491170 1153
where
~
C
ijlm
; ~g
qsck
and
~
D
ijck
are dimensionless elastic tensor, dimensionless viscoelastic compliance tensor and
the inverse of the dimensionless elastic tensor, respectively (see Appendix B).
In order to transform the stress rate into surface force rate it is necessary to impose the simplication
h
k
= h
l
= c only on boundary transformations. Therefore, expression (19) turns into
r
ij

E
ve
E
e
E
ve
E
e
_ _
~
C
lm
ij
e
lm

cE
ve
E
e
E
ve
E
e
_ _
~
C
lm
ij
_ e
lm

cE
ve
E
ve
E
e
_ _
_ r
ij
20
This is the desired rheological dierential relation for the Boltzmann model. More complicated rheological
models can be introduced following similar procedures, as the one employed in Appendix B, to achieve Eq.
(19). In Part I of the paper, for h
k
5h
l
, Eq. (19) is used changing only the last term to be equal to the last
term of Eq. (20), see Example 2.
Eq. (18) can be introduced into Eq. (3) resulting in an equilibrium equation similar to Eq. (15).
5. Integral equations and the BEM
As it is well known, the boundary element method is based on boundary integral equations. In this section
the divergence theorem is applied once to generate the starting displacement integral equation for both nite
and boundary elements [18,16]. After that the divergence theorem is applied again in order to obtain the
boundary integral equation considering the Kelvin and Boltzmann models, respectively.
The viscoelastic integral equation for boundary or internal points is obtained here using the weighting
residual technique in the dierential equilibrium equation (15) written in the following form:
r
a
ij;j
b
i
r
a
ij;j
r
ij;j
error
i
21
The error present in Eq. (21), when an approximate r
a
ij;j
solution is adopted, can be weighted by a proper
function. In this paper, the Kelvin fundamental solution for elastic innite body is adopted. Eq. (21) is
weighted over the analysed domain X, as follows:
_
X
u

ki
r
ij;j
b
i
dX 0 22
where u

ki
is the Kelvin fundamental solution. It represents the eect of a unit concentrated load applied to a
point located inside an innite domain. From Eq. (22) superscript a is omitted for simplicity. Applying the
divergence theorem to the rst term of Eq. (22), one achieves:
_
C
u

ki
r
ij
n
j
dC
_
X
u

ki;j
r
ij
dX
_
X
u

ki
b
i
dX 0 23
As previously mentioned, C is the boundary of the analysed body and n
j
is its outward normal vector. Know-
ing that r
ij
n
j
= p
i
, Eq. (4), and that u

ki;j
r
ij
e

kij
r
ij
, where e

kij
is the strain fundamental term, Eq. (23) turns
into:
_
C
u

ki
p
i
dC
_
X
e

kij
r
ij
dX
_
X
u

ki
b
i
dX 0 24
This equation is the starting point to obtain the viscoelastic integral representations for BEM and FEM con-
sidering both Kelvin and Boltzmann viscoelastic relations. In this part of the development the dierential vis-
cous relations, Eqs. (13) and (20), are used in the second term of Eq. (24) in order to derive a general BEM
formulation for viscoelastic analysis.
5.1. Displacement integral equation-Kelvin model
The boundary integral equation is developed replacing the total stress present in Eq. (24) by relation (6).
_
C
p
i
u

ki
dC
_
X
b
i
u

ki
dX
_
X
r
e
ij
e

kij
dX
_
X
r
v
ij
e

kij
dX 25
1154 A.D. Mesquita, H.B. Coda / Applied Mathematical Modelling 31 (2007) 11491170
Replacing r
e
ij
and r
v
ij
, given by Eqs. (7) and (8), in the last two terms of Eq. (25), gives:
_
C
p
i
u

ki
dC
_
X
b
i
u

ki
dX
_
X
e
lm
C
lm
ij
e

kij
dX
_
X
_ e
lm
g
lm
ij
e

kij
dX 26
At this point it is necessary to remember the kinematical relation for small strain, as follows:
e
ij

1
2
u
i;j
u
j;i
27
This relation can be derived with respect to time, resulting in the kinematical relation for strain velocity,
_ e
ij

1
2
_ u
i;j
_ u
j;i
28
Taking into account the stress and strain tensors symmetry, the kinematical relations and the symmetry of
C
ijlm
and g
ijlm
, one writes
e

kij
C
ijlm
e
lm
r

klm
e
lm
r

klm
u
l;m
r

kij
u
i;j
29
e

kij
g
ijlm
_ e
lm
e

kij
g
ijlm
_ e
lm
e

kij
g
ijlm
_ u
l;m
e

klm
g
lmij
_ u
i;j
30
By introducing these values into Eq. (26) and applying again the divergence theorem, the following integral
equation is found [16,17].
_
C
p
i
u

ki
dC
_
X
b
i
u

ki
dX
_
C
p

ki
u
i
dC
_
X
e

klm
g
lmij
_ u
i;j
dX
_
X
r

kij;j
u
i
dX 31
As usual in BEM formulations, due to the fundamental solution properties, the last integral of Eq. (31)
gives an independent term, leading to the following integral representation:
C
ki
u
i

_
C
p

ki
u
i
dC
_
X
e

klm
g
lmij
_ u
i;j
dX
_
C
p
i
u

ki
dC
_
X
b
i
u

ki
dX 32
where the free term C
ki
is equal to d
ki
(Kronecker delta) when the unit load is applied at internal points and
depends upon the geometry of the body if the collocation is chosen along the boundary [17,19].
The boundary of the analysed body, C, is discretized using n
e
boundary elements, C
e
. The domain is
approached by (n
c
) internal cells (X
c
), as shown in Fig. 6.
The variables are approximated by the following expressions:
u
i
n; t /
a
nU
a
i
t 33
p
i
n; t /
a
nP
a
i
t 34
b
i
n; g; t

/
b
n; gb
b
i
t 35
g
lmij
_ u
i;j
n; g; t g
lmij

/
b;j
n; g
_
U
b
i
t 36
where / and

/ are shape functions adopted for boundary and domain values, respectively and
U
a
i
; P
a
i
; b
b
i
and
_
U
b
i
are the nodal (b) displacement, surface force, body force and velocity values, respectively.

Cell Boundary
Element
Fig. 6. Problem discretization.
A.D. Mesquita, H.B. Coda / Applied Mathematical Modelling 31 (2007) 11491170 1155
Replacing these approximations into Eq. (32) results in
C
ki
u
i

_
C
e
p

ki
/
e
a
n dCU
a
i
t
_
X
e

klm
g
ij
lm

/
c
b;j
n; g dX
_
U
b
i
t
_
X
c
e

klm
g
lmij

/
c
b
;
j
n; g dX
_
U
b
i
t

_
C
e
u

ki
/
e
a
n dCP
a
i
t
_
X
c
u

ki

/
c
b
n dXb
b
i
t 37
where superscripts e and c represent boundary elements and internal cells, respectively.
Performing all spatial integrals and adopting the number of collocation points equal to the number of
nodes, the following system of time dierential equations is found:
HUt V
_
Ut GPt Bbt 38
where the new matrix V gives the viscosity contribution to the global movement; the other matrices and nodal
values are the same that appear in the standard static BEM approach [17,19].
The time dierential equation (38) can be integrated in time following a simple linear time approximation.
_
U
s1
U
s1
U
s
=Dt 39
where s + 1 represents the present instant and Dt is the adopted time step. From this approximation one writes
the time marching process:
HU
s1
GP
s1
F
s
40
where
H H
V
Dt
_ _
41
and
F
s
Bb
s1

V
Dt
U
s
42
5.2. Stress integral equation Kelvin model
As previously mentioned, to completely solve the problem it is necessary to calculate the stress eld for any
point of the analysed body. One can write the strain eld representation by replacing the displacement integral
representation, Eq. (32), into the kinematical relation (27), as follows:
e
ke
s; t
_
C
^e

kie
p
i
dC
_
C
^p

kie
u
i
dC
_
X
^e

kie
b
i
dX
_
X
^e

kije
g
ijlm
_ e
lm
dX ^ g

kije
g
ijlm
_ e
lm
s; t 43
where the derivatives of the kinematical relation were performed with respect to the source point (s)
co-ordinates.
The last two terms present in Eq. (43) come from the derivative of the last integral of the left side of Eq.
(32). Following standard textbooks on BEM [17,19], this derivative results in an integral in the sense of
Cauchy and a free term ^ g

kije
written together with the new fundamental values as follows:
^p

kie

1
4p1 mr
2
2m d
ki
r;
e
d
ei
r;
k
2d
ek
r;
i
8r;
k
r;
i
r;
e
f g
or
on
1 2m d
ki
n
e
d
ie
n
k
d
ke
n
i
2r;
k
r;
e
n
i

_
2mr;
i
r;
e
n
k
r;
i
r;
k
n
e

_
44
^e

kije

1
8p1 mGr
2
2m d
ki
r;
j
r;
e
d
kj
r;
i
r;
e
d
ei
r;
j
r;
k
d
ej
r;
i
r;
k
_ _
1 2m d
ki
d
je
d
kj
d
ie
_ _ _
2 d
ke
r;
i
r;
j
d
ij
r;
e
r;
k
4r;
k
r;
i
r;
j
r;
e
_ _
d
ij
d
ke

45
^ g

kije

1
81 tG
3 4td
ik
d
je

1
2
d
ke
d
ij

_ _
46
1156 A.D. Mesquita, H.B. Coda / Applied Mathematical Modelling 31 (2007) 11491170
The mentioned singular term of Eq. (43) is ^e

kije
g
lm
ij
_ e
lm
. From the integral representation of strains, one can
easily derive the elastic stress integral equation by applying the elastic constitutive relation (7), as follows:
r
e
kl
s
_
C
r

kil
p
i
dC
_
C
p

kil
u
i
dCdX
_
X
e

kijl
g
ijlm
_ e
lm
dX g
kijl
g
ijlm
_ e
lm
p
_
X
r

kil
b
i
dX 47
where the new fundamental values are given by
p

kil

2G
4p1 mr
2
2
or
on
1 2m d
kl
r;
i
m d
ki
r;
l
d
qi
r;
q
_ _
4r;
k
r;
i
r;
l
_ _
_
2m n
k
r;
l
r;
i
n
l
r;
k
r;
i
1 2m n
i
r;
k
r;
l
n
l
d
ki
n
k
d
li
1 4m n
i
d
kl
_
48
e

kijl

1
4p1 mr
2
1 2m d
ki
d
lj
d
ki
d
lj
d
kl
d
ji
2d
kl
r;
j
r;
i
_ _ _
2m d
kj
r;
l
r;
i
d
li
r;
k
r;
j
d
ki
r;
l
r;
j
d
lj
r;
k
r;
i
_ _
2d
ji
r;
k
r;
l
8r;
k
r;
i
r;
j
r;
l

49
g
kijl

1
81 m
8 6md
ki
d
lj
1 4md
kl
d
ij
50
Adding r
v
to both sides of Eq. (47) and taking into account Eqs. (8) and (6), the nal expression of the total
stress is
r
kl
s
_
C
r

kil
p
i
dC
_
C
p

kil
u
i
dCdX
_
X
e

kijl
g
ijlm
_ e
lm
dX g
kijl
g
ijlm
_ e
lm
p
_
X
r

kil
b
i
dX 51
where the new free term is given by
g
kijl

1
81 m
2d
ki
d
lj
1 4m d
kl
d
ij
52
As the total and elastic stresses are given by Eqs. (51) and (47), respectively, Eq. (6) must be adopted to
evaluate the viscous stress.
The same approximations, already dened to approximate the displacement equation, are adopted to
approach strain and stress integral equations. Thus, for any internal collocation point, strain, elastic stresses
and total stresses are expressed, respectively, by
e
kl
s; t
_
C
e
e

kil
/
a
dC
e
P
a
i
t
_
C
e
^p

kil
/
a
dC
e
U
a
i
t
_
X
c
^e

kijl
g
ijlm

/;
c
bm
dX
c
_
U
b
l
t

n
p
c1
^ g
kije
n
p
g
ijlm

/
b
;
m
s
_
U
b
l

_
X
e
e

kie
/
b
dX
c
b
b
i
53
r
e
kl
s; t
_
C
e
r

kil
/
a
dC
e
P
a
i
t
_
C
e
p

kil
/
a
dC
e
U
a
i
t
_
X
c
e

kijl
g
ijlm

/;
b
m
dX
c
_
U
b
l
t

n
p
c1
g
kije
n
p
g
ijlm

/
b
;
m
s
_
U
b
l

_
X
e
r

kie
/
b
dX
c
b
b
i
54
r
kl
s; t
_
C
e
r

kil
/
a
dC
e
P
a
i
t
_
C
e
p

kil
/
a
dC
e
U
a
i
t
_
X
c
e

kijl
g
ijlm

/;
b
m
dX
c
_
U
b
l
t

n
p
c1
g
kije
n
p
g
ijlm

/
b
;
m
s
_
U
b
l

_
X
e
r

kie
/
b
dX
c
b
b
i
55
where n
p
is the number of cells connected to the collocation point s. It means that, for this case, the free term is
calculated here as the average of the terms calculated for each common cell. After performing the integrals
over all boundary elements and internal cells and writing Eqs. (53)(55) for the same number of source points
as the number of discretization nodes [17,19], one has
A.D. Mesquita, H.B. Coda / Applied Mathematical Modelling 31 (2007) 11491170 1157
e
s1
G
0
P
s1
H
0
U
s1
B
0
b
s1
V
0
_
U
s1
56
r
s1
G
0
P
s1
H
0
U
s1
B
0
b
s1
V
0
_
U
s1
57
r
e
s1
G
0
P
s1
H
0
U
s1
B
0
b
s1
V
0
_
U
s1
58
where subscript s + 1 means that the analysed instant is t
s+1
= (s + 1)Dt. For simplicity, the presented BEM
formulation was derived for 2D plane strain problems. In plane stress situations one has to replace the Poisson
ratio m by m/(1 + m).
5.3. Displacement integral equation Boltzmann model
Eq. (24) is also the starting point for Boltzmann BEM viscoelastic formulation. Introducing Eq. (19) or (20)
into expression (24) one writes:
_
C
u

ki
p
i
dC
E
e
E
ve
E
e
E
ve
_
X
e

kij
~
C
ijlm
e
lm
dX
cE
e
E
ve
E
e
E
ve
_
X
e

kij
~
C
ijlm
_ e
lm
dX
cE
ve
E
e
E
ve
_
X
e

kij
_ r
ij
dX

_
X
u

ki
b
i
dX 0 59
Considering that e

kij
E
e
C
ijlm
e
lm
r

klm
e
lm
r

klm
u
l;m
r

kij
u
i;j
and e

kij
_ r
ij
u

ki;j
_ r
ij
, Eq. (59) is rewritten as
_
C
u

ki
p
i
dC
E
ve
E
e
E
ve
_
X
r

kij
u
i;j
dX
E
e
E
e
E
ve
_
X
e

kij
g
ijlm
_ e
lm
dX
cE
ve
E
e
E
ve
_
X
u

ki;j
_ r
ij
dX

_
X
u

ki
b
i
dX 0 60
Applying the divergence theorem to the second and fourth integrals of Eq. (60), results in
_
C
u

ki
p
i
dC
E
ve
E
e
E
ve
_
C
r

kij
n
j
u
i
dC
_
X
r

kij;j
u
i
dX
_ _

E
e
E
e
E
ve
_
X
e

kij
g
ijlm
_ e
lm
dX

cE
ve
E
e
E
ve
_
C
u

ki
_ r
ij
n
j
dC
_
X
u

ki
_ r
ij;j
dX
_ _

_
X
u

ki
b
i
dX 0 61
The third and the sixth integrals can be simplied by considering the dierential equilibrium equations of
the fundamental and real problems, i.e.:
r

kij;j
dp; sd
ki
62
_ r
ij;j

_
b
i
63
where
_
b
i
is the body force derived with respect to time and d(p, s) is the Diracs delta distribution. In Eq. (62) s
and p are, respectively, source and eld points.
Replacing Eqs. (62) and (63) into expression (61) and taking into account the Cauchy formula in its dier-
ential form _ r
ij
n
j
_ p
j
results,
C
ki
u
i
s
E
e
E
ve
E
ve
_
C
u

ki
p
i
dC
_
C
p

ki
u
i
dC
E
e
E
ve
_
X
e

kij
g
ijlm
_ e
lm
dX c
_
C
u

ki
_ p
i
dC
_
X
u

ki
_
b
i
dX
_ _

E
e
E
ve
E
ve
_
X
u

ki
b
i
dX 64
The free term C
ki
is the same that appears in the Kelvin viscoelastic model formulation. Eq. (64) is the inte-
gral representation for displacements based on Boltzmanns standard viscoelastic model. The rst, second and
sixth integrals are the same ones present in elastostatic problems [17,19].
In general the body force does not vary in time and is usually present before the analysis, therefore one may
consider
_
b
i
0, simplifying Eq. (64). The fourth integral presents no diculties, as the kernel singularity is the
1158 A.D. Mesquita, H.B. Coda / Applied Mathematical Modelling 31 (2007) 11491170
same as the rst one. Finally, the third integral is similar to the residual stress one present in non-linear anal-
ysis and presents only a weak singularity. The time marching process for the Boltzmann viscoelastic formu-
lation is described in Section 5.5.
5.4. Stress integral equation Boltzmann model
In order to achieve the desired integral representation, Eq. (64) is rewritten for internal points, as follows:
u
k
s
E
e
E
ve
E
ve
_
C
u

ki
p
i
dC
_
C
p

ki
u
i
dC
E
e
E
ve
_
X
e

kij
g
ijlm
_ e
lm
dX c
_
C
u

ki
_ p
i
dC
_
X
u

ki
_
b
i
dX
_ _

E
e
E
ve
E
ve
_
X
u

ki
b
i
dX 65
For small strain situation, the integral strain equation can be achieved applying the following kinematic
relation:
e
kl
s
1
2
ou
k
ox
l
s

ou
l
ox
k
s
_ _
66
The dierentiation expressed in Eq. (66) is dened with respect to the source point position. Applying Eq. (66)
to Eq. (65) and applying the Leibnitz formula to the third integral [17,19] one has
e
kl
s
E
e
E
ve
E
ve
_
C
^e

kil
p
i
dC
_
C
^p

kil
u
i
dC
E
e
E
ve
_
X
^e

kijl
g
ijlm
_ e
lm
dX ^ g
kijl
g
ijlm
_ e
lm
s
_ _
c
_
C
^e

kil
_ p
i
dC
_
X
^e

kil
_
b
i
dX
_ _

E
e
E
ve
E
ve
_
X
^e

kil
b
i
dX 67
The dierentiation is performed with respect to the source point, therefore only the fundamental values
change. Terms ^p

kil
; ^e

kijl
and ^ g

kijl
are the same previously presented for the Kelvin viscoelastic model, i.e.,
Eqs. (44)(46).
Rewriting the general viscoelastic relation, Eq. (20), as follows:
E
ve
E
e
E
ve
E
e
_ _
C
pqkl
e
kl
r
pq

cE
e
E
ve
E
e
_ _
~
C
pqkl
_ e
kl

cE
ve
E
ve
E
e
_ _
_ r
pq
68
and multiplying Eq. (67) by
E
ve
E
e
E
ve
E
e
C
pqkl
one achieves,
r
qq
s
_
C
r

qiq
p
i
dC
E
ve
E
e
E
ve
_
C
p

qiq
u
i
dC
E
e
E
e
E
ve
_
X
e

qijq
g
ijlm
_ e
lm
dX ~ g
qijq
g
ijlm
_ e
lm
s
_ _

cE
ve
E
e
E
ve
_
C
r

qiq
_ p
i
dC
_
X
r

qiq
_
b
i
dX
_ _

_
X
r

qiq
b
i
dX
cE
e
E
e
E
ve
C
qqlm
_ e
lm
s
cE
ve
E
e
E
ve
_ r
qq
s
69
The terms present in the rst bracket of Eq. (69) are an integral following the Cauchy principal sense and g
qijq
is a free term. Knowing that C
qqlm
_ e
ve
lm
d
qi
d
qj
C
ijlm
_ e
ve
lm
, it is possible to add the sixth term, on the right-hand side
of Eq. (69), to the free term ~ g
qijq
, resulting in
r
qq
s
_
C
r

qiq
p
i
dC
E
ve
E
e
E
ve
_
C
p

qiq
u
i
dC
E
e
E
e
E
ve
_
X
e

qijq
g
ijlm
_ e
lm
dX g
qijq
g
ijlm
_ e
lm
s
_ _

cE
ve
E
e
E
ve
_
C
r

qiq
_ p
i
dC
_
X
r

qiq
_
b
i
dX
_ _

_
X
r

qiq
b
i
dX
cE
ve
E
e
E
ve
_ r
qq
s 70
This is the total stress integral representation for the Boltzmann viscoelastic model. As previously mentioned,
the domain integral of strain rate should be performed following the Cauchy Principal Value sense. The new
fundamental values are the same given for the Kelvin viscoelastic model in Eqs. (48)(50).
A.D. Mesquita, H.B. Coda / Applied Mathematical Modelling 31 (2007) 11491170 1159
The evaluation of the elastic and viscous parts of the stress is performed at the approximation level, but in
order to prepare the necessary equation one can write expression (B1) in a time derivative form, i.e.:
_ r
el
ij
C
ijlm
_ e
ve
lm
E
ve
~
C
ijlm
_ e
ve
lm
71
Taking into account Eq. (B7), one transforms Eq. (71) as
r
v
ij
g
ijlm
C
ablm

1
_ r
el
ab
c
ijab
_ r
el
ab
or for h
l
h
k
cE
ve
C
ijlm
_ e
ve
lm
r
v
ij
c _ r
el
ij
72
From Eqs. (71), (72) and (18) one achieves
g
ijlm
C
ablm

1
_ r
el
ab
r
el
ij
r
ij
0 73
This equation is the basis of relaxation functions so widely used in the literature, which are achieved from
the supposition that the value of r
ij
is constant in a time step. In this paper, we propose a time approximation
for r
ij
and r
el
ij
that allows a more general and direct analysis of the global viscoelastic problem. This approx-
imation is described in details in the numerical time integration section.
For each boundary element and internal cell the variables are approached by parametric interpolations, as
follows:
p
i
/
a
P
a
i
74
u
i
/
a
U
a
i
75
b
i


/
a
b
a
i
76
_ p
i
/
a
_
P
a
i
77
_
b
i


/
a
_
b
a
i
78
g
ijlm
_ e
lm
g
ijlm
_ u
l;m
g
ijlm

/
a
0
m
_
U
a
l
79
where / and

/ are, respectively, the adopted shape functions for boundary element and internal cells. Indices a
and a are related to the boundary element and internal cell, respectively. Terms P
a
i
; U
a
i
; b
a
i
;
_
P
a
i
;
_
U
a
i
and
_
b
a
i
are
nodal values of surface forces, displacements, body forces and its time rates, respectively.
Substituting these approximations into integral Eqs. (64) and (70) results in
C
ki
U
i
s
E
e
E
ve
E
ve

n
e
e1
_
C
e
u

ki
/
a
dC
e
P
a
i

n
e
e1
_
C
e
p

ki
/
a
dC
e
U
a
i

E
e
E
ve

n
c
c1
_
X
c
e

kij
g
ijlm

/;
a
m
dX
c
_
U
a
l
c

n
e
e1
_
C
e
u

ki
/
a
dC
e
_
P
a
i
c

n
c
c1
_
X
c
u

ki

/
a
dX
c
_
b
a
i

E
e
E
ve
E
ve

n
c
c1
_
X
c
u

ki

/
a
dX
c
b
a
i
80
r
qq
s

n
e
e1
_
C
e
r

qiq
/
a
dC
e
P
a
i

E
ve
E
e
E
ve

n
e
e1
_
C
e
p

qiq
/
a
dC
e
U
a
i

E
e
E
e
E
ve

n
c
c1
_
X
c
e

qijq
g
ijlm

/;
a
m
dX
c
_
U
a
l

E
e
E
e
E
ve

n
p
c1
g
qijq
n
p
g
ijlm

/;
a
m
s
_
U
a
l

cE
ve
E
e
E
ve

n
e
e1
_
C
e
r

qiq
/
a
dC
e
_
P
a
i

cE
ve
E
e
E
ve

n
c
c1
_
X
c
r

qiq

/
a
dX
c
_
b
a
i

n
c
c1
_
X
c
r

qiq

/
a
dX
c
b
a
i

cE
ve
E
e
E
ve
_ r
qq
s 81
where n
p
is the number of cells that are connected to the source point s. It means that the free term is calculated
here as the average of the ones calculated for each common cell. After adopting the same number of source
points as the number of discretization nodes and performing the spatial integration [17,19], one has
HUt
E
e
E
ve
E
ve
GPt
E
e
E
ve
V
_
Ut cG
_
Pt cB
_
bt
E
e
E
ve
E
ve
Bbt 82
rt G
0
Pt
E
ve
E
e
E
ve
H
0
Ut
E
e
E
e
E
ve
V
0
_
Ut
cE
ve
E
e
E
ve
G
0
_
Pt
cE
ve
E
e
E
ve
B
0
_
bt B
0
bt
cE
ve
E
e
E
ve
_ rt 83
where t is time.
1160 A.D. Mesquita, H.B. Coda / Applied Mathematical Modelling 31 (2007) 11491170
5.5. Numerical time integration: (Boltzmann model)
For the Boltzmann model, a detailed explanation about the time integration procedure is necessary. Adopt-
ing linear time approximation generates the time marching process
_
U
s1

U
s1
U
s
Dt
84
_
P
s1

P
s1
P
s
Dt
85
_
b
s1

b
s1
b
s
Dt
86
_ r
s1

r
s1
r
s
Dt
87
_ r
el
s1

r
el
s1
r
el
s
Dt
88
where s + 1 stands for the instant t
s+1
.
Applying the above approximations to Eq. (82) one achieves the following system of algebraic equations:
HU
s1
GP
s1
F
s
89
where
G
c
Dt

E
e
E
ve
E
ve
_ _
G 90
H H
E
e
DtE
ve
V 91
F
s

E
e
DtE
ve
VU
s

c
Dt
GP
s
B
c
Dt

E
e
E
ve
E
ve
_ _
b
s1

c
Dt
b
s
_ _
92
Boundary conditions are imposed changing columns of H and G, as usually done in boundary element for-
mulations. As the values at instant t
s
are known, by solving Eq. (89) one nds P
s+1
and U
s+1
and is able to
solve the next time step and so on.
In order to calculate the total stress value it is necessary to evaluate
_
P
s1
;
_
U
s1
and
_
b
s1
, by means of Eqs.
(84)(86). From these values and Eq. (87) one calculates r
s+1
in Eq. (83), as follows:
r
s1
G
0
P
s1

E
ve
E
e
E
ve
H
0
U
s1

E
e
E
e
E
ve
V
0
_
U
s1

cE
ve
E
e
E
ve
G
0
_
P
s1

cE
ve
E
e
E
ve
B
0
_
b
s1
B
0
b
s1

c
Dt
E
ve
E
e
E
ve
r
s
_ _
_
1
c
Dt
E
ve
E
e
E
ve
_ _
93
or in a compact form
r
s1
G
0
P
s1
G
0
_
P
s1
H
0
U
s1


B
0
b
s1
V
0
_
U
s1
94
similarly to Eq. (57) including G
0
_
P
s1
.
Substituting approximation (88) into relation (73) it is possible to evaluate the elastic part of stress at the
Kelvin fragment of the adopted Boltzmann viscoelastic model, as follows:
r
el
ijs1
c
ijab

1
r
ijs1
r
el
ijs
_ _
or for h
l
h
k
r
el
s1
r
s1

c
Dt
r
el
s
_ __
1
c
Dt
_ _
95
where
c
ijab
c
ijab
Dt d
ijab
=Dt
r
el
ij
c
ijab
=Dtr
el
abs
96
where c
ijab
is dened in Eq. (72). The viscous stress r
v
s1
is calculated directly from Eq. (18) using the values of
r
s+1
and r
el
s1
determined above.
A.D. Mesquita, H.B. Coda / Applied Mathematical Modelling 31 (2007) 11491170 1161
5.6. Time marching processunied approach
In this item the solution procedure is summarized in order to clarify the expressions described in all items of
Section 5. Eqs. (40), (57), (58) and (89), (94) and (95) can be written in a unied form, as:
HU
s1
GP
s1
F
s
97
r
s1
G
0
P
s1
G
0
_
P
s1
H
0
U
s1


B
0
b
s1
V
0
_
U
s1
98
r
eg
s1
f r
eg
s1
99
where eg represents e for Kelvin model and el for Boltzmann model and f() is a simple recursive function.
As for the initial time t = 0 or s = 0 the restricted displacement, part of vector U
1
and the applied surface
force, part of vector P
1
, are known it is possible to calculate the unknown parts of displacement and surface
force vectors, for s = 0, using Eq. (97) as follows:

h
kk

h
ku

h
uk

h
uu
_ _
U
k
U
u
_ _
s1

g
uu
g
uk
g
ku
g
kk
_ _
P
u
P
k
_ _
s1

F
F
_ _
s1
100
where k is used for known values and u is used for unknown values. Isolating the unknown values on the left
side of the equation and the known ones on the right (changing columns) one writes
g
uu

h
ku
g
ku

h
uu
_ _
P
u
U
u
_ _
s1

h
kk
g
uk

h
uk
g
kk
_ _
U
k
P
k
_ _
s1

F
F
_ _
s

F
F
_ _
s1
101
or simply the resulting linear system of equations
AX
s1
F
s1
102
After knowing the complete vectors U
1
and P
1
one is able to achieve stresses directly from Eqs. (98) and
(99). From these values and the prescribed nodal surface forces and displacements for s = 1 the procedure
is repeated for any instant s, solving the problem.
6. Examples
In this section three examples are used to check the accuracy and stability of the method. The accuracy is
tested comparing numerical solutions with analytical ones. The stability is tested varying the adopted time step
for selected cases. Both Kelvin and Boltzmann model are applied to run the examples. For Kelvin model the
instantaneous Young modulus is not considered.
6.1. Simple stressed bar
The simple bar depicted in Fig. 7 is analysed when subjected to a force at its free end. As its analytical solu-
tion is very simple and easily obtained, this example is recommended to verify the accuracy of the proposed
P A
B
x
y
L
h
Physical Properties Geometry
E
e
= 22.5757 kN/mm
2
L = 800.0mm
E
ve
= 11.0 kN/mm
2
h = 100.0mm
= 0.0

= = = 45.454545
Analysis a Parameters Loading
t = 1.0 day P = 0.005 kN/mm
2
Number of time steps 450
Fig. 7. Geometry, discretizations and other required data.
1162 A.D. Mesquita, H.B. Coda / Applied Mathematical Modelling 31 (2007) 11491170
numerical model. The results obtained by analysing this simple bar adopting the proposed two-dimensional
boundary element technique (plane stress conditions were assumed) are compared with the one-dimensional
analytical solution. The boundary element discretization is given in Fig. 7. The physical constants adopted to
run this problem are also given in Fig. 7. For the two-dimensional analysis, the viscous parameters h
l
and h
k
are assumed to have the same value, taken here as equal to the one-dimensional case.
The free end displacements at point A, computed by BEM and given by the analytical solution, are depicted
in Fig. 8. As it can be seen, no dierences are observed when comparing the computed values. The stress values
computed at point B are given in Fig. 9. Again, no dierences among them are observed.
In Fig. 10 one can see that the present model is quite stable regarding time step length.
0 50 100 150 200 250 300 350 400 450
Time (days)
0.00
0.10
0.20
0.30
0.40
D
i
s
p
l
a
c
e
m
e
n
t

(
m
m
)
Analytic
BEM
0 50 100 150 200 250 300 350 400 450
Time (days)
0.10
0.20
0.30
0.40
0.50
0.60
D
i
s
p
l
a
c
e
m
e
n
t

(
m
m
)
Analytic
BEM
(a) Kelvin Model (b) Boltzmann Model
Fig. 8. Longitudinal displacements at point A.
0 50 100 150 200 250 300 350 400 450
Time (days)
0.000
0.001
0.002
0.003
0.004
0.005
S
t
r
e
s
s

(
k
N
/
m
m
2
)
Analytic (elastic)
Analytic (viscous)
BEM (total)
BEM (elastic)
BEM (viscous)
0 50 100 150 200 250 300 350 400 450
Time (days)
0.000
0.001
0.002
0.003
0.004
0.005
S
t
r
e
s
s

(
k
N
/
m
m
2
)
Analytic (elastic)
Analytic (viscous)
BEM (total)
BEM (elastic)
BEM (viscous)
(a) Kelvin Model (b) Boltzmann Model
Analytic (total)
Analytic (total)
Fig. 9. Stresses, r
e
11
; r
v
11
; r
11
, at point B.
A.D. Mesquita, H.B. Coda / Applied Mathematical Modelling 31 (2007) 11491170 1163
6.2. Internal pressure into a cylinder
A thick cylinder subjected to an internal pressure p is analysed. The double symmetry of the problem is
taken into account, therefore reducing the discretization of the body, as shown in Fig. 11. The geometric val-
ues and the physical parameters are also given in Fig. 11.
The example is organised in three cases: Case 1 h
k
= h
l
= c = 7.14285/day) is used to check the behaviour
of numerical solutions for both Kelvin and Boltzmann formulations. Case 2 (h
k
= 7.14285/day and h
l
= 0)
and Case 3 (h
l
= 7.14285/day and h
k
= 0) are used to check the extreme material behaviours regarding
non-proportional viscosity (using Kelvin formulation). Case one is compared with the analytical solution in
Figs. 12 and 13. The analytical solution is given when plane stress conditions are assumed.
ur r c
2
1 t
c
1
r
1 t
_ _
1
E
e

1
E
ve
1 e
t=c

_ _
where u is the radial displacement, r is the radius, c = h
l
= h
k
. Constants c
1
and c
2
depend on internal and
external pressures, p
i
and p
e
, respectively and are given by
P
y
x
x
y
R1
R2
Mechanical properties
E
ve
= 35.0 kgf/cm
2
E
e
=90.0 kgf/cm
2
= 0.4
Geometry
R
1
= 25.4cm
R
2
= 50.8cm
Other data

t = 1.0 day
number of t's =90
Load
P = 7.031 kgf/cm
2
Fig. 11. Geometry, discretization and other required data.
0 50 100 150 200 250 300 350 400 450
Time (days)
0.0
0.1
0.2
0.3
0.4
D
i
s
p
l
a
c
e
m
e
n
t

(
m
m
)
dt=0.5day
dt=1.0day
dt=5.0days
dt=10.0days
dt=25.0days
dt=50.0days
0 50 100 150 200 250 300 350 400 450
Time (days)
0.1
0.2
0.3
0.4
0.5
0.6
D
i
s
p
l
a
c
e
m
e
n
t

(
m
m
)
dt=0.5day
dt=1.0day
dt=5.0days
dt=10.0days
dt=25.0days
dt=50.0days
(a) Kelvin Model (b) Boltzmann Model
Fig. 10. Time step dependence.
1164 A.D. Mesquita, H.B. Coda / Applied Mathematical Modelling 31 (2007) 11491170
c
1

r
2
i
r
2
e
p
e
p
i

r
2
e
r
2
i
c
2

p
i
r
2
i
p
e
r
2
e
r
2
e
r
2
i
Again, for Kelvin formulation, one neglects the terms with E
e
, used only with the Boltzmann formulation.
The time step length dependence for the proposed scheme is also shown in Figs. 14 and 15. It is worth not-
ing that in those gures plane strain conditions are assumed in order to show the possibilities of the formu-
lation. As it can be observed, the obtained results are rather accurate even for large time steps. The nal
elastostatic solution is achieved for any chosen time step.
As previously mentioned, in order to describe the generality of the proposed procedure, the cylinder under
consideration is analysed for dierent values of h
l
and h
k
and Kelvin Model. Three situations are depicted in
0 20 40 60 80 100
Time (days)
0
2
4
6
8
10
12
D
i
s
p
l
a
c
e
m
e
n
t

(
c
m
)
Analytic
BEM
0 10 20 30 40 50 60 70 80 90
Time (days)
0
4
8
12
16
D
i
s
p
l
a
c
e
m
e
n
t

(
c
m
)
Analytic
BEM
(a) Kelvin Model (b) Boltzmann Model
Fig. 12. Inner wall displacements, comparison with analytical (plane stress).
0 10 20 30 40 50 60 70 80 90
Time (days)
0
2
4
6
8
D
i
s
p
l
a
c
e
m
e
n
t

(
c
m
)
Analytic
BEM
0 10 20 30 40 50 60 70 80 90
Time (days)
0
2
4
6
8
10
D
i
s
p
l
a
c
e
m
e
n
t

(
c
m
)
Analytic
BEM
(a) Kelvin Model (b) Boltzmann Model
Fig. 13. Outer wall displacements, comparison with analytical values (plane stress).
A.D. Mesquita, H.B. Coda / Applied Mathematical Modelling 31 (2007) 11491170 1165
Fig. 16. Case 1 means h
k
= h
l
= 7.14285, while h
k
= 7.14285 and h
l
= 0.0 dene Case 2 and h
k
= 0 and
h
l
= 7.14285 dene Case 3. From these gures, one can realise that h
l
is the most important coecient for
this problem in comparison with h
k
. The results are strongly dependent upon h
l
, while they practically do
not vary regarding h
k
. It means that if one needs to choose just one parameter for an analysis (Part II), it
should be h
l
.
6.3. Load changing
The same bar of the rst example is now analysed when subjected to a more general load. During the rst
200 time steps the load is sustained equal to 5 N/mm
2
. At this very instant it is changed to zero. The behaviour
0 10 20 30 40 50 60 70 80 90
Time (days)
0
2
4
6
8
10
D
i
s
p
l
a
c
e
m
e
n
t

(
c
m
)
dt=0.1day
dt=1day
dt=2.5days
dt=5days
dt=7.5days
dt=10days
dt=15days
0 10 20 30 40 50 60 70 80 90
Time (days)
0
2
4
6
8
10
12
14
D
i
s
p
l
a
c
e
m
e
n
t

(
c
m
)
dt=0.1day
dt=1day
dt=2.5days
dt=5days
dt=7.5days
dt=10days
dt=15days
(a) Kelvin Model (b) Boltzmann Model
Fig. 14. Inner wall radial displacements for several time steps (plane strain).
0 10 20 30 40 50 60 70 80 90
Time (days)
0
2
4
6
D
i
s
p
l
a
c
e
m
e
n
t

(
c
m
)
dt=0.1day
dt=1day
dt=2.5days
dt=5days
dt=7.5days
dt=10days
dt=15days
0 10 20 30 40 50 60 70 80 90
Time (days)
2
4
6
8
D
i
s
p
l
a
c
e
m
e
n
t

(
c
m
)
dt=0.1day
dt=1day
dt=2.5days
dt=5days
dt=7.5days
dt=10days
dt=15days
(a) Kelvin Model (b) Boltzmann Model
Fig. 15. Outer wall displacements for several time steps (plane strain).
1166 A.D. Mesquita, H.B. Coda / Applied Mathematical Modelling 31 (2007) 11491170
of the numerical solution is the same as the analytic one, as seen in Fig. 17. The Dt increment is the same used
in the rst example, i.e., one day. It has been assumed that h
k
= h
l
.
7. Conclusions and discussion
An alternative methodology for viscoelastic analysis has been presented using boundary element method.
This methodology is based on the global movement equation of solids and takes into account the non-local
characteristics of viscosity by continuity assumptions. The numerical implementation was performed for a
two-dimensional boundary element code using Kelvin and Boltzmann standard viscoelastic models. It was
demonstrated, by examples that no problems regarding convergence appear using the present formulation.
Accurate results were found for selected problems. Another advantage of this technique is that viscoelastic
problems are solved directly by a simple linear time domain scheme and not by residual decaying forces based
on relaxation functions, for which the statement of general boundary conditions are more complex. At the
0
2
4
6
8
10
12
Case 1
Case 2
Case 3
D
i
s
p
l
a
c
e
m
e
n
t

(
c
m
)
Time (day) Time (day)
0 20 40 60 80 100
0
1
2
3
4
5
6
7
Case 1
Case 2
Case 3
D
i
s
p
l
a
c
e
m
e
n
t

(
c
m
)
(a) Inner wall (b) Outer wall
0 20 40 60 80 100
Fig. 16. Dierent viscous parameters, Kelvin Model (plane stress).
0 100 200 300 400 500
Time (days)
0.00
0.05
0.10
0.15
0.20
0.25
0.30
0.35
0.40
D
i
s
p
l
a
c
e
m
e
n
t

(
m
m
)
Analytic
BEM
0
50 100 150 200 250 300 350 400 450
Time (days)
0.0
0.1
0.2
0.3
0.4
0.5
0.6
D
i
s
p
l
a
c
e
m
e
n
t

(
m
m
)
Analytic
BEM
(a) Kelvin Model (b) Boltzmann Model
Fig. 17. Displacement of point A for a changing.
A.D. Mesquita, H.B. Coda / Applied Mathematical Modelling 31 (2007) 11491170 1167
beginning of the developments it was dicult to nd a way to generate the Boltzmann model integral equa-
tion. A simplication regarding viscous parameters was necessary in order to write the stress rate as surface
force rate. This simplication does not invalidate the technique, but after hard considerations it opens the pos-
sibility of writing a boundary element formulation considering only boundary integrals. As a consequence the
use of cells can be avoided for both Kelvin and Boltzmann models, as shown in the next part of this study. One
should mention that the use of nite elements for the Boltzmann model considering h
k
5h
l
is recommended
when needed [19].
Appendix A. Additional remarks about the use of viscous dierential relations
The viscoelastic characteristics h
k
and h
l
can be identied from the results of a creep shear test (h
l
) and a
uniaxial tensile test (h
k
) [1]. The same procedure can be followed in order to identify the creep functions J(t)
and K(t).
One can relate the creep functions to the viscoelastic characteristics h
k
and h
l
, at any instant, by using the
following expressions:
h
l

t
ln1 2lJt Kt
A1
and
k1 2mh
k
2lh
l

Et
ln1 EJt
A2
for t 50 and, if desired, the values at t = 0, as follows:
h
l

1 2lJ0 K0
2ldJt=dt dKt=dt
A3
and
k1 2mh
k
2lh
l

1 EJ0
dJt=dt
A4
If one desires h
k
and h
l
to be constant, the curves, ln(1 2l(J(t) + K(t))) and ln(1 EJ(t)), should be
approximated by the best straight lines. If one assumes that h
k
and h
l
can vary along time, the above expres-
sions are directly applied and the proposed formulation does not change.
Appendix B. Boltzmann viscoelastic dierential relation
In this appendix some necessary steps to achieve expressions (16) and (17) from Eqs. (13)(15) are pre-
sented. It is important to show how to manage constitutive relations to achieve expressions that are more
general.
The elastic (instantaneous) part of the Boltzmann viscoelastic model is governed by the Hookes law, i.e.:
r
e
ij
C
ijlm
e
e
lm
E
e
~
C
ijlm
e
e
lm
B1
where C
lm
ij
is the usual elastic constitutive tensor, given by:
C
ijlm
kd
ij
d
lm
ld
il
d
jm
d
im
d
jl
B2
in which k and l are the Lame` constants, given in terms of Poisson ratio and Young modulus as:
k E
e
m
1 m1 2m
E
e
~
k B3
l E
e
1
21 m
E
e
~ l B4
~
C
ijlm

~
kd
ij
d
lm
~ ld
il
d
jm
d
im
d
jl
C
ijlm
=E
e
B5
where, for convenience, we dened new dimensionless values
~
k; ~ l and
~
C
ijlm
.
1168 A.D. Mesquita, H.B. Coda / Applied Mathematical Modelling 31 (2007) 11491170
Using the same idea, one writes the elastic part of the Kelvin viscoelastic stress as follows:
r
el
ij
C
ijlm
e
ve
lm
E
ve
~
C
ijlm
e
ve
lm
B6
The viscous stress is written in a general form as:
r
v
ij
g
ijlm
_ e
ve
lm
E
ve
~g
ijlm
_ e
ve
lm
E
ve
c
~
C
ijlm
_ e
ve
lm
B7
The viscous stress/strain rate relation is given by the tensor g
ijlm
, written as:
~g
ijlm
h
k
~
kd
ij
d
lm
h
l
~ ld
il
d
jm
d
im
d
jl
g
ijlm
=E
ve
B8
where, as for the Kelvin model, h
k
= h
l
are the viscous parameters.
The inverse representation of Eq. (B5) is:
~
D
ijlm

~
C
ijlm

1
2~ l
1
2
d
il
d
jm
d
im
d
jl

m
1 m
d
ij
d
lm
_ _
B9
Applying Eq. (B9) in Eqs. (B1) and (B6) and taking into account Eq. (13) one writes:
e
e
lm

1
E
e
~
D
lmij
r
ij
B10
e
ve
lm

1
E
ve
~
D
lmij
r
el
ij
B11
Form Eq. (15) one transforms Eq. (B11) into:
e
ve
lm

1
E
ve
~
D
lmij
r
ve
ij
r
v
ij

1
E
ve
~
D
lmij
r
ij
r
v
ij
B12
Assuming relation (B7) one writes Eq. (B12) as follows:
e
ve
lm

1
E
ve
~
D
lmij
r
ij

~
D
lmij
~g
ijck
_ e
ve
ck
B13
Substituting expressions (14) and (B10) into Eq. (B13) and rearranging terms, results:
e
lm

~
D
lmij
1
E
ve

1
E
e
_ _
r
ij

~
D
lmij
~g
ijck
_ e
ve
ck
B14
Multiplying Eq. (B14) by C
lm
qs
, given in Eq. (B5), results:
E
ve
E
e
E
ve
E
e
_ _
r
qs

~
C
qsck
e
ck
~g
qsck
_ e
ve
ck
B15
Dierentiating Eq. (14) with respect to time, one writes:
_ e
ve
ck
_ e
ck
_ e
e
ck
B16
Substituting relation (B16) into expression (B15) and rearranging indices, results:
E
ve
E
e
E
ve
E
e
_ _
r
qs

~
C
qsck
e
ck
~g
qsck
_ e
ck
~g
qsck
_ e
e
ck
B17
Dierentiating, with respect to time, Eq. (B10) and substituting into Eq. (B17) one achieves:
r
qs

E
ve
E
e
E
ve
E
e
_ _
~
C
qsck
e
ck

E
e
E
ve
E
ve
E
e
_ _
~g
qsck
_ e
ck

E
ve
E
ve
E
e
_ _
~g
qsij
~
D
ijck
_ r
ck
B18
In order to transform the stress rate into surface force rate it is necessary to impose the simplication
h
k
= h
l
= c. In this way expression (B18) turns into:
r
ij

E
ve
E
e
E
ve
E
e
_ _
~
C
ijlm
e
lm

cE
ve
E
e
E
ve
E
e
_ _
~
C
ijlm
_ e
lm

cE
ve
E
ve
E
e
_ _
_ r
ij
B19
A.D. Mesquita, H.B. Coda / Applied Mathematical Modelling 31 (2007) 11491170 1169
This is the general rheological dierential relation for the Boltzmann model. It is employed to derive the
boundary element procedure proposed here. More complicate rheological models can be introduced following
similar procedures as the employed to achieve Eq. (B19).
References
[1] J. Lemaitre, J.L. Chaboche, Mechanics of Solids, Cambridge University Press, Cambridge, 1990.
[2] W. Flu gge, Viscoelasticity, Blaisdell Publication, USA, 1967.
[3] Z. Sobotka, Rheology of Materials and Engineering Structures, Elsevier, Prague, Czechoslovakia, 1984.
[4] R.M. Christensen, Theory of Viscoelasticity, Academic Press, New York, 1982.
[5] W.H. Chen, C.M. Chang, J.T. Yeh, An incremental relaxation nite element analysis of viscoelastic problems with contact and
friction, Comput. Meth. Appl. Mech. Eng. 109 (1993) 315319.
[6] W.C. Carpenter, Viscoelastic stress analysis, Int. J. Numer. Meth. Eng. 4 (1972) 357366.
[7] W.H. Chen, T.C. Lin, Dynamic analysis of viscoelastic structures using incremental nite element method, Eng. Struct. 14 (1982) 271
276.
[8] S. Yadagiri, R.C. Papi, Viscoelastic analysis of near incompressible solids, Comput. Struct. 20 (1985) 817825.
[9] J. Argyris, I.St. Doltsinis, V.D. Silva, Constitutive modelling and computation of non-linear viscoelastic solids. Part I: Rheological
models and integration techniques, Comp. Meth. Appl. Mech. Eng. 88 (1991) 135163.
[10] P. Perzyna, The constitutive equations for rate sensitive plastic models, Quart. Appl. Math. 20 (1963) 321332.
[11] D.R.J. Owen, F. Damjanic, Viscoplastic analysis of solids stability considerations, Recent Advances in Non-Linear Computational
Mechanics, Pineridge Press, Swansea, 1982.
[12] J. Munaiar Neto, A study of viscoelastic and viscoplastic constitutive models and the use of Finite Elements algorithms, Doctor in
Science Thesis, EESC-USP, Brazil, 1998.
[13] J.H. Argyris, I.S.T. Doltsinis, K.J. William, New developments in the inelastic analysis of quasistatic and dynamic problems, Int. J.
Numer. Meth. Eng. 14 (1979) 18131850.
[14] J.C. Simo, T.J.R. Hughes, Computational Inelasticity, Springer, New York, 1998.
[15] I. Doltsinis, Elements of Plasticity Theory and Computation, WIT Press, Southampton, 2000.
[16] A.D. Mesquita, H.B. Coda, W.S. Venturini, An alternative time marching process for viscoelastic analysis by BEM and FEM, Inter.
J. Numer. Meth. Eng. 51 (2001) 11571173.
[17] C.A. Brebbia, J.C.F. Telles, L.C. Wrobel, Boundary Element Techniques: Theory and Applications in Engineering, Springer, Berlin,
1984.
[18] A.D. Mesquita, H.B. Coda, Alternative Kelvin viscoelastic procedure for nite elements, Appl. Math. Model. 26/4 (2002) 501516.
[19] G. Beer, J.O. Watson, Introduction to Finite and Boundary Element Methods for Engineers, Wiley, New York, 1992.
1170 A.D. Mesquita, H.B. Coda / Applied Mathematical Modelling 31 (2007) 11491170