Comput Mech

DOI 10.1007/s00466-011-0667-5
ORIGINAL PAPER
Variational multiscale approach to enforce perfect bond
in multiple-point constraint applications when forming
composite beams
R. Emre Erkmen Mark A. Bradford Keith Crews
Received: 30 June 2011 / Accepted: 16 November 2011
Springer-Verlag 2011
Abstract Composite laminates that consist of two or more
layers nd widespread applications in a variety of engineer-
ing structures. In the computational modelling of composite
laminates, the layers can be stacked together and connected
conveniently at the nodes by using multiple-point constraints
(MPCs). However, this type of modelling leads to weakening
of the kinematic constraint conditions imposed by the bond
between the juxtaposed layers and as a consequence, MPCs
application at the nodes produces behaviour that is softer than
the perfectly bonded composite beam behaviour. The work
herein shows that when kinematic conditions for composite
action are weakly imposed in the variational form, they can
be enforced in the point-wise sense by proper selection of
the interpolation eld or otherwise reinforced by using vari-
ational multiscale approach without modifying the kinematic
model. The originality of the approach presented herein is in
the interpretation of the MPCs application as the solution
in a superuously extended space because of the weaken-
ing in the kinematic constraints. It is shown that the perfect
bond between the composite beam layers can be recovered
by excluding the identied ne-scale effect from the solu-
tion of the multiple point constraint application. The conver-
gence characteristic of the nite element formulation is also
improved by using the variational multi-scale approach. It
is also shown that the ne-scale effects can be represented
R. E. Erkmen (B) K. Crews
School of Civil and Environmental Engineering,
University of Technology, Sydney, NSW 2007, Australia
e-mail: emre.erkmen@uts.edu.au
K. Crews
e-mail: k.crews@uts.edu.au
M. A. Bradford
School of Civil and Environmental Engineering, The University
of New South Wales, Sydney, NSW 2052, Australia
e-mail: m.bradford@unsw.edu.au
by using extra ctitious elements and springs, which offers a
direct correction technique in modelling of composite beams
that is especially useful when access to the numerical proce-
dure is limited.
Keywords Variational multiscale method
Composite beams Multiple-point constraints
Perfect bond Interpolation error
1 Introduction
Composite laminates that consist of two or more different
layers stacked together nd widespread applications in a
variety of engineering structures. Numerous laminate the-
ories have been proposed hitherto to describe the kinemat-
ics and stress states of composite laminates. The classical
theory for these assumes perfect bond between the layers,
for which Kirchhoff kinematics is used [1]. In the present
study, the analysis is limited to composite beams based on
the kinematic assumptions of the classical theory. A prac-
tical application in the modelling of composite beams is to
use Multiple-Point Constraints (MPCs) at the nodes to bond
the layers together. However, this type of modelling does
not inherit the kinematic behaviour of the continuous case
and thus full-interaction between the layers cannot be always
imposed by applying MPCs at the nodes. Gupta and Ma [2]
pointed out this fact and noted that the source of error in MPC
applications of this type can be related to the incompatibil-
ity in the displacement eld. A similar type of error in MPC
applications for built-up plates and shells was pointed out by
Criseld [3].
On the other hand, an interpolated displacement eld can
be conceived as a displacement eld of an extended interpo-
lation space under a constraint condition [4]. Since the MPC
123
Comput Mech
application imposes a weaker condition than the perfectly
bonded case, the interpolation space in the MPC application
is treated herein as a superuously extended space. By using
the variational multiscale approach introduced by Hughes et
al. [57], a nite element formulation based on the perfect
bond between the layers is recovered by excluding the ne-
scale effect from the solution of the MPC application. Thus,
the targeted solution is initially the coarse-scale solution for
the problem considered herein. The convergence character-
istic of the nite element formulation based on the perfectly
bonded case is also improved by using the variational mul-
tiscale approach in a standard manner. The improvements
in the accuracy and convergence characteristics are illus-
trated. It is also shown that the effects of the ne-scale eld
can be represented by using extra ctitious elements and
springs, which offers a direct correction technique that is
especially useful when the access to the numerical procedure
is limited.
The paper is organised as follows. The kinematics and the
weak form of the equilibrium equations for the composite
beam layers are introduced in Sect. 2. In Sect. 3, nite ele-
ment formulations are developed for composite beam analy-
sis by using MPCs and alternatively by enforcing the perfect
bond between the layers as a-priori condition. In Sect. 4, it is
shown that by using the variational multiscale approach the
nite element formulation based on the perfect bond can be
recovered from the formulation based on MPC application.
In Sect. 5, a nite element formulation that provides exact
values at the nodes is obtained also using the variational mul-
tiscale approach. Numerical examples are presentedinSect. 6
and conclusions are drawn in Sect. 7.
2 Composite beam kinematics and nite element
solution
2.1 Displacements and strains
The composite beam is made up of n layers in which an
arbitrary layer is referred to as layer i , i.e., i = 1, 2, . . . n.
According to Kirchhoff kinematics, the deformations of a
layer can be expressed in terms of the axial displacements
w
i
and the vertical displacements v
i
of their centroids. Thus,
the strain expression in each layer can be determined in terms
of the axial displacement gradients Dw
i
, and the curvatures
D
2
v
i
due to bending as

i
= Dw
i
y
i
D
2
v
i
(1)
in which
i
is the strain in layer i, y
i
is the distance of a point
on the cross-section from the centroid of layer i and D( ) =
d( )/dz, where z refers to the axial coordinate of the beam.
2.2 Weak form of the equilibrium equations
A displacement-based nite element formulation can be
developed by employing the principle of virtual work, i.e.
=
n

i =1
_
L
_
A
1

i
dAdz
ext
= 0, (2)
where the rst integral is the virtual work done due to the
deformations of the layers and
ext
is the virtual work done
by the external forces. In Eq. 2,
i
is the stress conguration
in layer i , which in general can be related with the strain
conguration
i
through a secant stiffness modulus, i.e.
i
=
E
i

i
, A
i
, is the cross-sectional area, andL is the span of the
analysis domain. Routinely, by substituting Eq.1 into Eq. 2,
the weak form of equilibrium equations can be written as
_
L
u
T
(z)Du(z)dz
_
L
u
T
(z)q(z)dz = 0, (3)
where D is the matrix of the cross-sectional properties, i.e.,
D =

E
1
A
1
0 0 0 0 0
0 E
1
I
1
0 0 . . . 0 0

0 0 E
i
A
i
0 0
0 0 0 E
i
I
i
0 0

0 0 0 0 E
n
A
n
0
0 0 0 0 0 E
n
I
n

, (4)
in which I
i
=
_
A
i
y
2
i
dA is the moment of inertia of the cross-
section of layer i with respect to the bending axis passing
through the centroid of layer i . It is assumed that a single
secant stiffness modulus E
i
can be used throughout the layer.
In Eq. 4, the displacement vector u can be written in terms
of the axial and vertical displacement components of each
layer as
u(z)=
_
w
1
(z) v
1
(z) w
i
(z) v
i
(z) w
n
(z) v
n
(z)
_
T
, (5)
while the deformation vector u can be written as
u(z) =
_
Dw
1
(z)D
2
v
1
(z) Dw
i
(z)D
2
v
i
(z)
Dw
n
(z)D
2
v
n
(z)
_
T
. (6)
In Eq. 3, q(z) is the vector of external actions, i.e.,
q(z) = q
w1
(z) q
v1
(z) q
wi
(z) q
vi
(z)
q
wn
(z) q
vn
(z)
T
. (7)
The composite action is enforced in the next section by using
two different approaches; MPC application at the nodes and
alternatively enforcing perfect bond kinematics in the point-
wise sense.
123
Comput Mech
3 Finite element solution to enforce composite action
3.1 MPC application to enforce composite action
To develop a nite element formulation, by using Eq. 3 the
variational form of the equilibrium equations can be written
as
d
T

_
L
N
T
(z)B
T
DBN(z)dz

d d
T
f = 0, (8)
where
N(z) =

w
1
(z) 0 0 0 0 0
0 v
1
(z) 0 0 0 0

0 0 w
i
(z) 0 0 0
0 0 0 v
i
(z) 0 0

0 0 0 0 w
n
(z) 0
0 0 0 0 0 v
n
(z)

,
(9)
and
B =

D 0 0 0 0 0
0 D
2
0 0 0 0

0 0 D 0 0
0 0 0 D
2
0 0

0 0 0 0 D 0
0 0 0 0 0 D
2

. (10)
In Eq. 8, f is the energy equivalent nodal force vector, i.e.,
f =
_
L
N
T
(z)q(z)dz, (11)
and d is the vector of nodal displacements associated with
the selected interpolation functions including the nods for all
layers, e.g., w
i
(z) and v
i
(z) used in Eq. 9, from which the
displacement eld used in Eq. 8 can be expressed as
u(z) = N(z)d, (12)
and deformation eld can be expressed as
u(z) = BN(z)d. (13)
When layers are bonded together according to Kirchhoff
kinematics, vertical separation between the layers should be
prevented and the composite cross-section should remain
planar after the deformation. In the nite element mod-
elling of composites, separate layers can be connected
convenientlyat the nodes byusingMPCs. Inthis case, assum-
ing that the nodes of the mth layer are the master nodes,
nodal displacements of the other layers can be obtained by
using the no vertical separation rule, i.e. v
k
(0) = v
m
(0)
and v
k
(L) = v
m
(L), and the plane sections remain plane
rule at the nodes i.e., w
k
(0) = w
m
(0) h
k
Dv
m
(0) and
w
k
(L) = w
m
(L) h
k
Dv
m
(L), where h
k
= y
k
y
m
is
the difference between the coordinates of the centroids of
layers k and m, i.e. k = 1, 2, . . . n and k = m. It should
be noted that throughout the document index k is used to
indicate any element except the element of the master nodes
m, on the other hand index i refers to any element including
the element of the master nodes m. By using the introduced
nodal constraint conditions in Eq. 8, variational form of the
equilibrium equations can be obtained as

d
T

d
T

f = 0, (14)
in which

K = T
T

_
L
N
T
(z)B
T
DBN(z)dz

T (15)
and

f = T
T
_
L
N
T
(z)q(z)dz, (16)
where T is the matrix of nodal constraint conditions and

d is
the vector of retained nodal degrees of freedom, i.e. d = T

d
and

f = T
T
f. Matrix T is obtained in the following.
3.2 Stiffness matrix due to MPC application
The basic nite element can be developed by using linear
interpolations for the axial displacements and cubic interpo-
lation for the vertical displacements, i.e.,
w
i
(z) =
_
(1 z/L) z/L
_
, (17)
and
v
i
(z) =
__
1
3z
2
L
2
+
2z
3
L
3
__
z
2z
2
L
+
z
3
L
2
_
_
3z
2
L
2

2z
3
L
3
__

z
2
L
+
z
3
L
2
__
. (18)
The nodal displacement vector in this case can be written as
d
T
=
_
w
T
N 1
v
T
N 1
w
T
N i
v
T
N i
w
T
N n
v
T
N n
_
T
, (19)
in which w
T
Ni
= w
i
(0)w
i
(L), and v
T
Ni
= v
i
(0)v

i
(0)v
i
(L)v

i
(L). In this case, the matrix of nodal constraint condi-
tions can be written as
T =
_
T
1
T
k
T
m1
I
m
T
m+1
T
n
_
T
, (20)
123
Comput Mech
Fig. 1 Flow of the relationships
where
T
k
=

1 0 0 0 0 0
0 1 0 0 0 0
0 0 1 0 0 0
h
k
0 0 1 0 0
0 0 0 0 1 0
0 h
k
0 0 0 1

, (21)
I
m
=

1 0 0 0 0 0
0 1 0 0 0 0
0 0 1 0 0 0
0 0 0 1 0 0
0 0 0 0 1 0
0 0 0 0 0 1

, (22)
and the vector of retained nodal degrees of freedom which is
those of the m
th
layer can be written as

d
T
=
_
w
T
N m
v
T
N m
_
. (23)
By substituting Eqs. 4, 6, 9, 17 and 18 into Eq. 15, the stiff-
ness matrix can be obtained as

K =
_

K
11

K
12

K
T
12

K
22
_
, (24)
in which

K
11
=
1
L

i =1
E
i
A
i

n

i =1
E
i
A
i

i =1
E
i
A
i
n

i =1
E
i
A
i

, (25)

K
12
=
1
L

0
n

i =1
E
i
A
i
h
i
0
n

i =1
E
i
A
i
h
i
0
n

i =1
E
i
A
i
h
i
0
n

i =1
E
i
A
i
h
i

(26)
and

K
22
=
n

i =1
E
i
I
i
A +
n

i =1
E
i
A
i
h
2
i
S, (27)
where
A =
1
L
3

12 6L 12 6L
6L 4L
2
6L 2L
2
12 6L 12 6L
6L 2L
2
6L 4L
2

, (28)
and
S =
1
L

0 0 0 0
0 1 0 1
0 0 0 0
0 1 0 1

. (29)
This solution will be referred herein as Multiple-Point Con-
straint Solution (MPCS). It will be shown, however, that the
imposition of the MPCs at the nodes is decient in that it
does not inherit the kinematic properties of the continuous
problem introduced in the next section. However, the de-
ciency in MPCS can be remedied by using the variational
multiscale method as will be shown in Sect. 4 (Fig.1).
3.3 Finite element solution satisfying kinematic constraints
in the point-wise sense
When two layers are juxtaposed, in order to enforce perfect
bond between the layers (i.e., full interaction in the point-
wise sense), the kinematic conditions can be modied prior
to imposing nite element interpolation functions. By impos-
ing the kinematic constraint conditions of no vertical sepa-
ration (i.e., v
k
= v
m
) and the cross-section remains planar
after the deformation (i.e., w
k
= w
m
hDv
m
) in Eq. 2, the
variational form of the equilibrium equations becomes
_
L

_
Dw
m
D
2
v
m
_

i =1
E
i
A
i

n

i =1
E
i
A
i
h
i

i =1
E
i
A
i
h
i
n

i =1
E
i
I
i
+
n

i =1
E
i
A
i
h
2
i

_
Dw
m
D
2
v
m
_

dz
ext
= 0. (30)
123
Comput Mech
Similarly to the MPCS in Sect. 3.1, a nite element formu-
lation can be developed by using linear interpolation for the
axial displacement eld w
m
and cubic interpolation for the
vertical displacement eld v
m
, i.e.

d
T
K

d
T

f = 0, (31)
in which
K =
_
L

N
T
(z)

B
T

T
T
D

N(z)dz, (32)
and

f =
_
L

N
T
(z)

t
T
q(z)dz, (33)
where

f is the energy equivalent nodal force vector. Matrices

B and

N(z) can be written as

B =
_
D 0
0 D
2
_
, (34)

N(z) =
_
w
m
(z) 0
0 v
m
(z)
_
, (35)
Matrix

T that imposes kinematic constraints on the continu-
ous deformations as used in Eq. 32 can be explicitly written
as

T =
_

T
1
. . .

T
k
. . .

T
m1

I
m

T
m+1
. . .

T
n
_
T
, (36)
in which

T
k
=
_
1 0
h
k
1
_
, (37)
and

I
m
=
_
1 0
0 1
_
, (38)
and matrix

t that imposes kinematic constraints on the con-
tinuous deections as used in Eq. 33 can be written as

t =
_

t
1
. . .

t
k
. . .

t
m1

I
m

t
m+1
. . .

t
n
_
T
(39)
in which

t
k
=
_
1 0
h
k
D 1
_
. (40)
By substituting Eqs. 4, 6, 9, 17 and 18 into Eq. 15, the stiff-
ness matrix can be obtained as
K =

K
11
K
12
K
T
12
K
22

, (41)
where K
11
=

K
11
, K
12
=

K
12
and K
22
can be explicitly
written as
K
22
=
_
n

i =1
E
i
I
i
+
n

i =1
E
i
A
i
h
2
i
_
A. (42)
This solution will be referred herein as Perfect-Bond Kine-
matics Finite element Solution (PBKFS).
4 Enforcement of perfect bond using the variational
multiscale approach
4.1 Variational multiscale approach
In this section, variational multiscale approach will be
adopted to produce the nite element formulation PBKFS
as introduced in Sect. 3.3, using the MPC application pro-
cedure as introduced in Sect. 3.2. This approach avoids
a-priori modication of the kinematics of the problem to
enforce perfect bond between the layers. In the variational
multiscale approach [5], the displacement eld vector u is
decomposed into coarse and ne-scale displacement elds,
u and u

respectively, i.e., u = u + u

. The ne-scale dis-

placement eld vanishes at the element boundaries, i.e.,
u

(0) = u

(L) = 0 and the spaces of the course- and

ne-scale functions are linearly independent. Likewise, the
deformation eld is decomposed herein into two linearly
independent components, i.e. u = u + u

where the
coarse-scale deformation eld is u =

T

N(z)

d, in which

T,

B and

N(z) are as shown in Eqs. 3436, respectively. The
ne-scale displacement eld can in general be expressed as
u

= N

(z)d

, where N

denotes the vector of ne-scale inter-

polation functions such as bubble functions or p-renements
[8], and d

denotes the vector of ne-scale nodal parameters.

From the ne-scale displacement eld, the ne-scale defor-
mation eld can be obtained as u

= BN

(z)d

. The aimis to
recover the nite element formulation based on perfect bond
between the composite beam layers in the point-wise sense
by excluding the ne-scale effect fromthe nite element for-
mulation based on MPC application. For this purpose, Eq.
14 is split into two interacting forms by using the variational
multiscale approach, i.e.

d
T
K

d +

d
T

d
T

f = 0 (43)
and
d

K
T

d + d

T
K

T
f

= 0, (44)
where
K

=
_
L
N

T
(z)B
T
DBN

(z)dz (45)
123
Comput Mech
and

=
_
L

N
T
(z)

B
T

T
T
DBN

(z)dz. (46)
It should be noted that in obtaining Eqs. 43 and 44 from
Eq. 14, an additive split for the weighting functions was also
introduced, i.e. u = u + u

. In between the above two

interacting equations, the ne-scale nodal parameters, i.e. d

in Eq. 43 can be eliminated in order to be left with a problem

entirely in terms of the coarse-scale solution. The elimination
of d

can be achieved by using the usual static condensation,

and thus fromEq. 44 the vector of ne scale nodal parameters
d

can be calculated as
d

= K

1
_
f

d
_
. (47)
By substituting Eq. 47 into Eq. 43, the equilibrium equa-
tions that consider the effect of the ne-scale solution can be
written as

d =

f, (48)
From the results based on Eq. 14, by substituting into Eq. 48
the coarse scale solution can be extracted as
K =

K +

K

T
(49)
and

f =

f +

K

1
f

, (50)
where
f

=
_
L
N
T
(z)q(z)dz. (51)
It should be noted that in contrast to the conventional appli-
cation of the variational multiscale method, initially the
required solution is the coarse-scale solution in Eqs. 49 and
50. In the next section, the interpolation space of the ne-
scale displacement eld is identied.
4.2 Fine- scale displacement eld
The displacement eld used in MPCS, i.e. u = NT

d, is based
on linear interpolations of the axial displacements w
i
, and
cubic interpolations of the vertical displacements v
i
. On the
other hand, the displacement eld in PBKFS, i.e. u =

t

N

d,
is based on a quadratic axial displacement eld for w
k
due
to the contribution of the cubic vertical displacement eld
v
m
, i.e. w
k
= w
m
hDv
m
. Considering this difference in
the interpolation spaces of the two solutions the ne-scale
interpolation functions can be determined as
N

(z) =

T
1
(z)

N

T
k
(z)

N

T
m1
(z)
0
N

T
m+1
(z)

N

T
n
(z)

,
(52)
where non-diagonal terms are zero,
N

k
(z) =
_
z
L
_
1
z
L
_
0
_
, (53)
and
0
T
= 0 0 . (54)
Despite the fact that it contains higher order terms PBKFS
is identied herein as the coarse-scale solution, which may
seem initially counter-intuitive. Therefore, in order to clar-
ify why PBKFS is the coarse scale solution, an argument
based on the hierarchical enrichment procedure and static
condensation of the associated hierarchical degrees-of-free-
dom [6] is presented in the following. For this purpose,
it is crucial to note that in PBKFS the quadratic bubble
is enforced without having its own associated hierarchical
degree-of-freedom, which prevents PBKFS from attaining
the optimal solution considering the space of all interpola-
tions of the axial displacement elds w
k
up to quadratic order.
Comparison to PBKFS, MPCS is the optimal solution in the
sense that when the axial displacement elds w
k
are enriched
using quadratic bubble functions with associated hierarchi-
cal degrees-of-freedom, i.e. w
k
(z) = w
k
(z)w
T
N
k
+ N

k
(z)b,
degrees-of-freedomassociated with the quadratic bubble, i.e.
will not be activated. This means, MPCS is the optimal
solution within the space of all interpolations of the axial
displacement elds w
k
up to quadratic order. When the extra
degrees-of-freedom, i.e. b is condensed out, there will be no
change in the MPCS stiffness matrix

K. This is because there
is no coupling between the hierarchical degrees of freedom
and those based on the linear interpolation w
T
N
k
, i.e.
_
L
_
dw
k
(z)
dz
w
T
N
k
E
k
A
k
dN

k
(z)
dz
b
_
dz
= E
k
A
k
_
_
w
k
(L) w
k
(0)
L
_

_
z
L

z
2
L
2
_

L
0
b
_
= 0
(55)
Among those solutions where multiple-point constraints
are applied to connect composite beam layers, PBKFS is
123
Comput Mech
prevented from attaining the optimal solution due to further
enforcement of the perfect bond in the point-wise sense, and
therefore it is considered herein as the coarse-scale solution.
5 Exact solution at the nodes
5.1 Differential equations and interpolation functions
From Eq. 30, by using integration by parts, the variational
formof the differential equilibriumequations canbe obtained
as
_
L
_ _
w
m
v
m
_

i =1
E
i
A
i
D
2

i =1
E
i
A
i
h
i
D
3

i =1
E
i
A
i
h
i
D
3
_
n

i =1
E
i
I
i
+
n

i =1
E
i
A
i
h
2
i
D
4
_

_
w
m
v
m
__
dz
ext

Boun
|
L
0
= 0. (56)
The boundary conditions are satised at z =0 and z =L, thus
the last term in Eq. 56, i.e.
Boun
|
L
0
vanishes. The solution
of the differential equations in Eq. 56 can be obtained as
_
w
m
(z)
v
m
(z)
_
=
_
w
mh
(z)
v
mh
(z)
_
+
_
w
mp
(z)
v
mp
(z)
_
= N
e
(z)

d + u
p
(z),
(57)
in which the rst and second terms are the homogenous and
particular solutions, respectively. In Eq. 57, N
e
(z).
N
e
(z) =

w
m
(z)
n

i =1
E
i
A
i
h
i
n

i =1
E
i
A
i
_
Dv
m
(z) M

(z)
_
0 v
m
(z)

(58)
in which
M

(z) = 0 (1 z/l) 0 z/L . (59)

By using the homogenous solution of Eq. 56 as the interpola-
tion eld of the nite element formulation, and thus replacing

N(z) with N
e
(z) in Eq. 32, the stiffness matrix of the nite
element can be obtained as
K
e
=
_
L
N
T
e
(z)

B
T

T
T
D

BN
e
(z) dz. (60)
where
K
e
=

K
11
K
12
K
T
12
K
e22

. (61)
Comparison to the matrix components in Eq. 41, the dif-
ference is only due to the sub-matrix K
e22
which can be
explicitly written as
K
e22
=

i =1
E
i
I
i
+
n

i =1
E
i
A
i
h
2
i

_
n

i =1
E
i
A
i
h
i
_
2
n

i =1
E
i
A
i

A
+

_
n

i =1
E
i
A
i
h
i
_
2
n

i =1
E
i
A
i

i =1
E
i
A
i
h
2
i

S.
(62)
On the other hand, by using Eq. 58, the energy equivalent
nodal force vector can be obtained as
f
e
=
_
L
N
T
e
(z)

t
T
q(z) dz, (63)
This solution will be referred herein as Perfect-Bond Kine-
matics Exact element Solution (PBKES). It should be noted
that regardless of the load vector q(z) , the nodal displace-
ment values obtained by using K
e
and f
e
are exact. This is
because the effect of particular solution on the weak form of
equilibrium equations vanishes [9]. In order to illustrate that,
the particular solution u
p
in Eq. 57 can be expressed as
u
p
= + N
e
(z)

d
p
. (64)
where the components of the displacement vector
T
=

w
. consists of any displacement function satisfying the
equilibrium equations in the presence of member load terms,
i.e. non-homogenous case. In Eq. 64, the second displace-
ment eld vector consists of any linear combination of the
homogeneous solution. Thus, if vector is a particular solu-
tion for the differential equations of equilibrium, vector. u
p
is also a particular solution irrespective of the magnitude of
the arbitrary constants in vector.

d
p
. Conventionally, once a
particular solution is determined, the entries of vector

d
p
are
selected to vanish for simplicity. In contrast, we deviate from
this convention herein by selecting the entries

d
p
in order for
the particular solution vanish at the boundaries, i.e.
v
mp
(0) = Dv
mp
(0) = w
mp
(0) = v
mp
(L)
= Dv
mp
(L) = w
mp
(L) = 0 (65)
The vector of displacement derivatives can be written in
terms of the homogeneous and particular parts as
(z) = (z) +

BN
e
(z)
_

d +

d
p
_
. (66)
in which =
_
D
w
D
2
v
_
T
and
T
=
_
D
w
D
2

v
_
. It should be
noted that the variation of the prescribed components of the
displacement vector in Eq. 66 vanishes, i.e. =

BN
e

d
p
.
Thus, for the general non-homogeneous loading case, the
virtual work expression can be written as
=

d
T
K
e

d

d
T
f
e
+
e
= 0. (67)
123
Comput Mech
in which
e
can be written as

e
=
_
L
_
Dw
mh
D
2
v
mh
_

i =1
E
i
A
i

n

i =1
E
i
A
i
h
i

i =1
E
i
A
i
h
i
n

i =1
E
i
I
i
+
n

i =1
E
i
A
i
h
2
i

_
Dw
mp
D
2
v
mp
_
dz (68)
By integrating the terms including Dw
mp
once by parts, and
those including D
2
v
mp
twice by parts
e
can be written as

e
=
_
L
_
w
mp
v
mp
_

i =1
E
i
A
i
D
2
w
mh

n

i =1
E
i
A
i
h
i
D
3
v
mh

i =1
E
i
A
i
h
i
D
3
w
mh
+
_
n

i =1
E
i
I
i
+
n

i =1
E
i
A
i
h
2
i
_
D
4
v
mh

dz

i =1
E
i
A
i
h
i
D
2
w
mh

i =1
E
i
I
i
+
n

i =1
E
i
A
i
h
2
i

D
3
v
mh

Dv
mp

L
0
+

i =1
E
i
A
i
h
i
D
3
w
mh

i =1
E
i
I
i
+
n

i =1
E
i
A
i
h
2
i

D
2
v
mh

v
mp

L
0
+

i =1
E
i
A
i
D
2
w
mh

n

i =1
E
i
A
i
h
i
D
3
v
mh

w
mp

L
0
(69)
in which the displacement elds v
mh
and w
mh
satisfy the
homogenousformof theequilibriumequationsandv
mp
, Dv
mp
and w
mp
satisfy the boundary conditions in Eq. 65, thus both
the eld equations and boundary conditions vanish in Eq. 69.
5.2 Variational multiscale approach to obtain exact
element stiffness matrix and the load vector
By using the variational multi-scale approach it can be shown
that PBKFS can be modied to obtain the stiffness matrix
and the energy equivalent load vector of PBKES which is an
improved solution due to enriched interpolation space how-
ever, based on the same perfect-bond kinematics that is sat-
ised in the point-wise sense. Considering the interpolation
spaces of PBKFS which is based on linear interpolations
for the axial displacement eld w
m
, and cubic interpolations
for the vertical displacements eld v
m
and PBKES which
is based on the quadratic interpolation of the axial displace-
ment eld w
m
, and cubic interpolations of the vertical dis-
placements eld v
m
, the vector of the ne-scale interpolation
functions can be determined as
N

e
(z) =
_
z
L
_
1
z
L
_
0
_
(70)
Using N

e
(z) in Eq. 70, it can be veried that the stiffness
matrix K
e
and the energy equivalent load vector of PBKES
given in Eqs. 61 and 63, respectively can be obtained as
K
e
= K

K

e
K

1
e

K
e

T
(71)
and
f
e
=

f

K

e
K

1
e
f

e
, (72)
in which
K

e
=
_
L
N

T
e
(z)

B
T

T
T
D

BN

e
(z)dz, (73)

e
=
_
L

N
T
(z)

B
T

T
T
D

BN

e
(z)dz, (74)
and
f

e
=
_
L
N

T
e
(z)

t
T
q(z)dz, (75)
were used. It is noted that in this study PBKFS is consid-
ered as the coarse-scale solution in both kinematic models,
i.e. the model in Sect. 3.1 where separate layers are bonded
together using multiple-point constraints at the nodes, and
the model in Sect. 3.3 where perfect bond between the layers
is enforced as a-priori condition. Originality of this study
is in the identication of the perfectly-bonded nite ele-
ment solution PBKFS as the coarse scale solution within
the rst model. On the other hand, within the context of
the second model, PBKFS is the basic nite element for-
mulation and PBKES is the improved solution that contains
ne-scale effects due to enriched interpolation space with
quadratic bubble. Modifying PBKFS using variational mul-
tiscale method in order to obtain the results of PBKES is
according to the standard definitions of the coarse and ne-
scale solutions e.g., [6]. In order to further clarify the dif-
ferences between MPCS and PBKES it should be noted that
when a nite element formulation is developed according
to the rst model, i.e. in MPCS axial displacement elds of
all n layers are interpolated, whereas according to the sec-
ond model, i.e. in PBKES only one axial displacement eld
is interpolated. The ow of the relationships is presented in
Fig. 1.
123
Comput Mech
Fig. 2 Cantilever beam with
two layers
Master node
Slave node
Master node
Slave node
E
1
=200x10
3
MPa A
1
=6x10
3
mm
2
I
1
=112.2x10
3
mm
4
E
2
=26x10
3
MPa A
2
=7.1x10
3
mm
2
I
2
=124x10
6
mm
4
EI
n
=4.28x10
12
Nmm
2
k
spr
=-2.14x10
9
Nmm
Fictitious spring Fictitious element
h=163mm
L=2000mm
6 Applications
6.1 Effect of loading on the numerical error
In order to illustrate the numerical error in MPCS solution
and the effect of loading on the error, a 2 m span compos-
ite cantilever beam with two layers is analysed. As shown
in Fig. 2, the top component of the composite beam has a
modulus of elasticity of E
1
= 200 10
3
MPa, cross-sec-
tional area of A
1
= 6 10
3
mm
2
and the moment of inertia
of I
1
= 112.2 10
3
mm
4
. The bottom component has a
modulus of elasticity of E
2
= 26 10
3
MPa, cross-sectional
area of A
2
= 7.1 10
3
mm
2
and the moment of inertia of
I
2
= 124 10
6
mm
4
. The distance between the centroids of
the two layers is 163 mm. For the multiple-point constraint
application the master nodes are selected as the nodes of the
bottom component and thus the nodes of the top component
are slave nodes, i.e. h = 163 mm.
6.1.1 Beam under uniform bending moment
Under uniform internal bending moment acting on the beam
due to a 100 kNm moment applied at the tip, the vertical
deection diagramis as shown in Fig. 3a and the axial deec-
tions at the centroids of bottom and top layers are as shown
in Fig. 3b and c respectively. From Fig. 3, it can be observed
that both the MPCSand PBKFSare in perfect agreement with
PBKES, which is the exact solution as also illustrated in [2].
Thus, the conventional MPCS solution does not require any
correction when the composite beam is under pure bending
behaviour and thus, perfect bond between the layers is satis-
ed in the point-wise sense.
6.1.2 Beam under uniform shear
Under a vertical tip load of 50 kN which causes a uniform
internal shear force distribution along the beam, the verti-
cal and the axial deection at the centroids of the layers are
shown in Fig. 4. In this case, internal shear causes differences
in the solutions. FromFig. 4a it can be observed that one-ele-
ment MPCS is softer than both PBKFS and PBKES, which
is due to incompatibility [2]; On the other hand, because of
the interpolation errors the nite element behaviour based
on PBKFS is stiffer than PBKES, which is the exact solu-
tion in this case. When the number elements are increased
0
10
20
30
0 500 1000 1500 2000
MPCS 1 element
PBKFS 1 element
PBKES 1 element
0
1
2
3
4
0 500 1000 1500 2000
MPCS 1 element
PBKFS 1 element
PBKES 1 element
A
x
i
a
l

d
e
f
l
e
c
t
i
o
n

b
o
t
t
o
m

l
a
y
e
r

(
m
m
)

Beam span (mm)
-0.6
-0.5
-0.4
-0.3
-0.2
-0.1
0
0 500 1000 1500 2000
MPCS 1 element
PBKFS 1 element
PBKES 1 element
Beam span (mm)
A
x
i
a
l

d
e
f
l
e
c
t
i
o
n

t
o
p

l
a
y
e
r

(
m
m
)
V
e
r
t
i
c
a
l

d
e
f
l
e
c
t
i
o
n

(
m
m
)
Beam span (mm)
a
b
c
Fig. 3 a Vertical deections under uniform bending moment. b Axial
deection of the bottom layer under uniform bending moment. c Axial
deection of the top layer under uniform bending moment
to four the interpolation errors reduce significantly and the
results based on both MPCS and PBKFS converge to the
exact solution. As shown in Fig. 4b, axial deections at
123
Comput Mech
0
10
20
30
0 500 1000 1500 2000
MPCS 1 element
MPCS 4 element
PBKFS 1 element
PBKFS 4 element
PBKES 1 element
V
e
r
t
i
c
a
l

d
e
f
l
e
c
t
i
o
n

(
m
m
)
Beam span (mm)
0
0.5
1
1.5
2
0 500 1000 1500 2000
MPCS 1 element
MPCS 4 element
PBKFS 1 element
PBKFS 4 element
PBKES 1 element A
x
i
a
l

d
e
f
l
e
c
t
i
o
n

b
o
t
t
o
m

l
a
y
e
r

(
m
m
)
Beam span (mm)
-0.4
-0.3
-0.2
-0.1
0
0 500 1000 1500 2000
MPCS 1 element
MPCS 4 element
PBKFS 1 element
PBKFS 4 element
PBKES 1 element
A
x
i
a
l

d
e
f
l
e
c
t
i
o
n

t
o
p

l
a
y
e
r

(
m
m
)
Beam span (mm)
a
b
c
Fig. 4 a Vertical deections under vertical tip load. b Axial deection
of the bottom layer under vertical tip load. c Axial deection of the top
layer under vertical tip load
the bottomlayer based on MPCS and PBKFS are both piece-
wise linear whereas for PBKES it is a parabolic function. On
the other hand, it is shown in Fig. 4c that the axial deection
due to PBKFS solution is parabolic whereas the axial deec-
tion based on MPCS is still linear. The accuracies of the
numerical solutions are evaluated based on the normalised
error according to the strain energy of the exact solution.
The convergence rate (i.e. p in e
en
= Ca
p
where C is an
arbitrary constant and a is the element size) can be obtained
from the slope of the log-log error curve shown in Fig. 5,
-10
-8
-6
-4
-2
0
-3 -2 -1 0
L
o
g
(
e
n
e
r
g
y

e
r
r
o
r
)
Log
2
(element length/span)
MPC
PBKFS
2
Fig. 5 Convergence rate of the conventional element under vertical tip
load
from which it can be veried that both the accuracy and con-
vergence rate of PBKFS are better than those of MPCS. The
slope for MPCS is 9.4, whereas the slope for PBKFS is 10.5.
6.1.3 Beam under uniformly distributed vertical load
Numerical characteristics of the solutions are similar to the
previous case under a uniformly distributed vertical load
of 100kN/m. Vertical deections in this case are shown in
Fig. 6a from which it can be veried that again MPCS is
softer and PBKFS is stiffer than the exact solution. It is also
shown that based on PBKES one- and four-element solutions
provide the same nodal displacements, thus nodal values are
exact. Fig. 6b and c shows the axial displacements at the
centroids of the bottom and top layers, respectively. Com-
parison to the previous case, the accuracies of the solutions
based on both MPCS and PBKFS reduce as can be observed
from Fig. 7, which is normalized according to eighty ele-
ment PBKES solution. From the slope of Fig. 7, the rate of
the convergence for MPCS and PBKFS can be determined
as 7.7 and 8.8 respectively. Thus, it can be veried that the
convergence rates also reduce in comparison to the previous
case.
6.2 Fine-scale effect on the stiffness matrix and the load
vector
For instance for the problem described in Sect. 6.1, the dif-
ference in the stiffness matrices between MPCS and PBKES
can be expressed using the variational multiscale approach
as
K
e


K
22
=

0 0
0
_
E
1
A
1
E
2
A
2
h
2
E
1
A
1
+E
2
A
2
_
A
_
E
1
A
1
E
2
A
2
h
2
E
1
A
1
+E
2
A
2
_
S

. (76)
The error due to the lacking terms in the stiffness matrix
of MPCS can be compensated by introducing an additional
ctitious beam element and a ctitious spring between the
123
Comput Mech
0
10
20
30
40
0 500 1000 1500 2000
MPCS 1 element
MPCS 4 element
PBKFS 1 element
PBKFS 4 element
PBKES 1 element
PBKES 4 element
V
e
r
t
i
c
a
l

d
e
f
l
e
c
t
i
o
n

(
m
m
)
Beam span (mm)
0
1
2
3
0 500 1000 1500 2000
MPCS 1 element
MPCS 4 element
PBKFS 1 element
PBKFS 4 element
PBKES 1 element
PBKES 4 element
A
x
i
a
l

d
e
f
l
e
c
t
i
o
n

b
o
t
t
o
m

l
a
y
e
r

(
m
m
)
Beam span (mm)
-0.5
-0.4
-0.3
-0.2
-0.1
0
0 500 1000 1500 2000
MPCS 1 element
MPCS 4 element
PBKFS 1 element
PBKFS 4 element
PBKES 1 element
PBKES 4 element
A
x
i
a
l

d
e
f
l
e
c
t
i
o
n

t
o
p

l
a
y
e
r

(
m
m
)
Beam span (mm)
a
b
c
Fig. 6 a Vertical deections under vertical uniform distributed load.
b Axial deection of the bottom layer under vertical uniform distrib-
uted load. c Axial deection of the top layer under vertical uniform
distributed load
master nodes as shown in Fig. 2. Use of ctitious elements
with negative stiffnesses in order to be able to adopt the avail-
able numerical procedure can be found in the literature e.g.,
[1012]. Consideringthe differences inthe stiffness terms the
spring coefcient that should be used to correct the model
can be determined as
k
spr
=
1
L
_
E
1
A
1
E
2
A
2
h
2
E
1
A
1
+ E
2
A
2
_
, (77)
-8
-6
-4
-2
0
-3 -2 -1 0
L
o
g
2
(
e
n
e
r
g
y

e
r
r
o
r
)
Log
2
(element length/span)
MPC
PBKFS
Fig. 7 Convergence rate of the conventional element under vertical
uniform distributed load
and the bending rigidity of the ctitious element can be deter-
mined as
EI
n
=
_
E
1
A
1
E
2
A
2
h
2
E
1
A
1
+ E
2
A
2
_
. (78)
6.3 Effect of stiffness of layers on the numerical error
In order to illustrate the effect of relative layer stiffness on
the numerical error a 2000mm span three layer simply sup-
ported composite beam (Fig. 8) is analysed. Cross-sectional
areas and moment of inertias of the layers are taken as A =
5 10
3
mm
2
and I
1
= 100 10
3
mm
4
, respectively. Total
height of the cross-section is 150mm and each layer has
equal depth and thus the distance between the centroids of
layers is 50mm. In the MPCS model, the nodal points are
at the centroid for each layer and the nodes of the bottom
layer are selected as the master nodes on the cross-section.
The modulus of elasticity for the middle layer is taken as
E
1
= 200 10
3
MPa, and for the top and bottom layers the
modulus of elasticity, i.e. E
2
, is varied for parametric stud-
ies. The proportion of the moduli of elasticity of the layers is
denoted with , i.e. = E
1
/E
2
. In Fig. 9a, error in the strain
energies of two- and eight-element MPCS solution are plot-
ted for 0.1 10. Although the error reduce for when
top and bottom layers get relatively softer (e.g., in the case
of elasto-plastic deformations) error in two-element MPCS
solution is dramatic. The error could be kept below 2% by
only using 80 elements in the MPCSmodel and the results are
not effected when the layer of the master nodes are changed
since the interpolation functions for all the layers are identi-
cal. It should be noted that for pure bending case (i.e. when
equal and opposite bending moments are applied at the ends),
the error in the strain energy of MPCS model is zero. On the
other hand, PBKFS provides much better results for two-
and eight-element solutions as shown in Fig. 9b. However,
PBKFS is affected by the selection of the layer of the master
nodes and when it is changed to middle layer the error in
123
Comput Mech
P
Master node
Slave node
Master node
Slave node
E
2
E
1
E
2
Fig. 8 Simply supported composite beam with three layers
-3000%
-2500%
-2000%
-1500%
-1000%
-500%
0%
0 2 4 6 8 10
E
r
r
o
r

i
n

t
o
t
a
l

s
t
r
a
i
n

e
n
e
r
g
y

MPCS-2elements
MPCS-8elements
0%
10%
20%
30%
0 2 4 6 8 10
E
r
r
o
r

i
n

t
o
t
a
l

s
t
r
a
i
n

e
n
e
r
g
y

PBKFS-2elements
PBKFS-8elements
a
b
Fig. 9 a Effect of relative layer stiffness on the error in MPCS. b Effect
of relative layer stiffness on the error in PBKFS
PBKFS solution is zero since linear interpolation of the axial
displacement eld is exact.
7 Conclusions
In this study, it was shown that variational multiscale
approach can be an adequate numerical paradigmfor applica-
tions of multiple-point constraints to impose full composite
action. When composite beams are formed using multiple-
point constraints at the nodes, the weakening of the bond in
between the composite beamlayers, and thus the overly ex-
ible behaviour comparison to the behaviour of the perfectly
bonded case, was considered as being due to the superu-
ous extension of the interpolation space. By considering this
extension in the interpolation space, the numerical solution
for the perfectly bonded composite beam could be recovered
within the frame work of the variational multiscale approach
from the results of the multiple-point constraint application
without modifying the kinematic model of the nite element
formulation. By using the ne-scale effects, a correction pro-
cedure to obtain exact results using ctitious elements and
springs was also presented. Effects of external load types and
relative layer stiffness on the numerical error in multiple-
point constraints applications were illustrated.
References
1. Reddy JN (1993) An evaluation of equivalent-single-layer and
layer-wise theories of composite laminates. Compos Struct 25:21
35
2. Gupta AK, Paul SM (1977) Error in eccentric beam formulation.
Int J Numer Method Eng 11:14731483
3. Criseld MA (1991) The eccentricity issue in the design of plate
and shell elements. Commun Appl Numer Method 7:4756
4. Courant R, Hilbert D (1962) Methods of mathematical physics.
Wiley, New York
5. Hughes TJR (1995) Multiscale phenomena: Greens functions, the
Dirichlet-to-Neumann formulation, subgrid scale models, bubbles
and the origins of stabilized methods. Comput Method Appl Mech
Eng 127:387401
6. Hughes TJR, Feijoo GR, Mazzei L, Quincy J-B (1998) The varia-
tional multiscale method: a paradigmfor computational mechanics.
Comput Method Appl Mech Eng 166:324
7. Hughes TJR, Sangalli G (2007) Variational multiscale method:
The ne-scale Greens function, projection, optimization, locali-
zation, and stabilized methods. SIAM J Numer Anal 45:539557
8. Brezzi F, Franca LP, Hughes TJR, Russo A (1997) b =
_
g. Com-
put Method Appl Mech Eng 145:329339
9. Erkmen RE, Bradford MA (2011) Treatment of slip-locking for
displacement based nite element analysis of composite beam-
columns. Int J Numer Method Eng 85:805826
10. Miles JC (1976) The determination of collapse load and energy
absorbing properties of thin walled beam structures using matrix
methods of analysis. Int J Mech Sci 18:399402
11. Rutenberg A (1981) Adirect P-delta analysis using standard plane
frame computer programs. Computers and Structures 14:97102
12. Challamel N, Pijaudier-Cabot (2006) Stability and dynamics of a
plastic softening oscillator. Int J Solids Struct 43:58675885
123