You are on page 1of 9

Journal of Computational and Applied Mathematics 168 (2004) 383391

www.elsevier.com/locate/cam
A dual CraigBampton method for dynamic substructuring
Daniel J. Rixen
;1
T.U. Delft, Faculty of Design, Engineering and Production, Engineering MechanicsDynamics,
Mekelweg 2, Delft 2628 CD, The Netherlands
Received 19 July 2002
Abstract
A novel component mode synthesis method for dynamic analysis of structures is presented. It is based
on free interface vibration modes and residual exibility components. Although the ingredients are the same
as in previously published procedures (e.g. MacNeal or Rubin), our method is fundamentally dierent in
that it assembles the substructures using interface forces (dual assembly) and enforces only weak interface
compatibility. The new formulation is based on a fully consistent reduction approach and the reduced matrices
so-obtained are exactly dual to the CraigBampton reduced matrices. The new free interface substructuring
method proposed here is thus more natural then classical free interface synthesis procedures and leads to
simpler reduced matrices. We illustrate the eciency of the dual CraigBampton approach for reducing the
dynamical representation on a three-dimensional frame.
c 2003 Elsevier B.V. All rights reserved.
Keywords: Component mode synthesis; Dynamic substructuring; FETI; Model reduction; Free interface mode
1. Introduction
The power and storage capability of modern computing hardware as well as remarkable advances
in numerical methods and in compilers make it possible nowadays to solve very large linear systems
(typically of the order of millions of degrees of freedom). Nevertheless, since dynamic analysis (e.g.
inverse iteration based eigensolvers or time-integration schemes) requires solving many linear systems
and because the complexity and renement of nite element models is increasing at least as fast as
the computing capabilities, dynamic substructuring remains an essential tool for engineers. Building
reduced models of subparts (known as super- or macro-elements in nite elements) facilitates sharing

Tel.: 3115-278-1523; fax: 3115-278-2150.


E-mail address: d.j.rixen@wbmt.tudelft.nl (D.J. Rixen).
1
Supported by the Koiter Institute, Delft University of Technology.
0377-0427/$ - see front matter c 2003 Elsevier B.V. All rights reserved.
doi:10.1016/j.cam.2003.12.014
384 D.J. Rixen / Journal of Computational and Applied Mathematics 168 (2004) 383391
models between design groups. Reducing the representation of the dynamical behavior of subparts
is also very relevant for building reduced order systems and designing high performance controllers.
For an overview of substructuring methods see for instance [1].
Basically, there are two classes of mode components synthesis methods: those where the substruc-
ture dynamics is represented by xed interface vibration modes and those where free interface modes
are used. The method commonly used is the CraigBampton procedure [2]. Nearly all substructuring
methods reduce the subparts in sub-domains with displacement connectors on the interface so that
they can be used in nite element codes as macro-elements.
In this paper, we will present a method that is based on free interface modes and where the
substructures are assembled through interface forces. We will see that this combination is fully
natural and leads to an attractive dual CraigBampton formulation.
In Section 2, we discuss the dierence between primal and dual assembly. In Section 3, we then
develop the dual CraigBampton method. A numerical example is given in Section 4.
2. Primal and dual assembly of substructures
Let us assume that a nite element modal dened on a domain is subdivided into a number
N
(s)
of substructures called
(s)
such that every node belongs to one and only one substructure
except for those on the interface boundaries. The linear dynamic behavior of each substructure
(s)
is governed by the local equilibrium equations
M
(s)
u
(s)
+K
(s)
u
(s)
=f
(s)
+g
(s)
; s = 1; : : : ; N
s
; (1)
where M
(s)
and K
(s)
are the substructure mass and stiness matrices, u
(s)
are the local degrees
of freedom, f
(s)
the external loads applied to the substructure and g
(s)
the internal forces on the
interfaces between substructures that ensure compatibility.
2.1. Primal assembly of substructures
Every substructure can be looked at as a macro-element: the local degrees of freedom u
(s)
are
related to a global set of assembled degrees of freedom by
u
(s)
=

u
(s)
b
u
(s)
i

L
(s)
b
0
0 I

u
b
u
(s)
i

; (2)
where the subscripts i and b refer, respectively, to the internal and boundary degrees of freedom and
where L
(s)
b
is a Boolean localization matrix relating assembled degrees of freedom u
b
on the interface
to u
(s)
b
. Using the compatibility condition (2), the local equilibrium Eq. (1) can be assembled as
M
a
u
a
+K
a
u
a
=f
a
(t); (3)
D.J. Rixen / Journal of Computational and Applied Mathematics 168 (2004) 383391 385
where
M
a
=

N
s

s=1
L
(s)
T
b
M
(s)
bb
L
(s)
b
L
(1)
T
b
M
(1)
bi
L
(N
s
)
T
b
M
(N
s
)
bi
M
(1)
ib
L
(1)
b
M
(1)
ii
0
.
.
.
.
.
.
M
(N
s
)
ib
L
(N
s
)
b
0 M
(N
s
)
ii

; u
a
=

u
b
u
(1)
i
.
.
.
u
(N
s
)
i

;
K
a
=

N
s

s=1
L
(s)
T
b
K
(s)
bb
L
(s)
b
L
(1)
T
b
K
(1)
bi
L
(N
s
)
T
b
K
(N
s
)
bi
K
(1)
ib
L
(1)
b
K
(1)
ii
0
.
.
.
.
.
.
K
(N
s
)
ib
L
(N
s
)
b
0 K
(N
s
)
ii

(4)
and where u
a
is the second-order time derivative of u
a
. The interface forces g
(s)
cancel out when
assembled on the interface.
2.2. Dual assembly of substructures
Another way to enforce the compatibility between substructures consists in explicitly expressing
the compatibility constraints, namely the equality between corresponding degrees of freedom on the
interface
M
(s)
u
(s)
+K
(s)
u
(s)
+

b
(s)
T
z
0

=f
(s)
;
N
s

s=1
b
(s)
u
(s)
b
= 0;
(5)
where b
(s)
are signed Boolean matrices. Comparing with (1), b
(s)
T
z represent the interconnecting
forces between substructures. We introduce the block-diagonal notations
M =

M
(1)
0
.
.
.
0 M
(N
s
)

; K =

K
(1)
0
.
.
.
0 K
(N
s
)

;
386 D.J. Rixen / Journal of Computational and Applied Mathematics 168 (2004) 383391
u =

u
(1)
.
.
.
u
(N
s
)

; f =

f
(1)
.
.
.
f
(N
s
)

;
B = [ B
(1)
B
(N
s
)
]; B
(s)
T
=

b
(s)
T
0

: (6)
The set of Eqs. (5) can then be written in block form

M 0
0 0

u
z

K B
T
B 0

u
z

f
0

: (7)
The dual assembled system (7) is equivalent to the assembled system (3) since they express the
same local equilibrium and enforce the same interface compatibility.
3. The dual CraigBampton method
3.1. CraigBampton
According to the primal assembled system (3), every substructure can be looked at as being
excited through its interface degrees of freedom, namely
M
(s)
ii
u
(s)
i
+K
(s)
ii
u
(s)
i
=f
(s)
i
K
(s)
ib
u
(s)
b
M
(s)
ib
u
(s)
b
; (8)
which indicates that one can dene an approximation for u
(s)
i
in every subdomain as a superposition
of a static response and of eigenmodes associated to M
(s)
ii
and K
(s)
ii
. Hence, one can dene a
transformation matrix for reducing the primal assembled system as
u
a
=

u
b
u
(1)
i
.
.
.
u
(N
s
)
i

T
CB

u
b
q
(1)
.
.
.
q
(N
s
)

I 0 0
T
(1)
L
(1)
b
4
(1)
0
.
.
.
.
.
.
T
(N
s
)
L
(N
s
)
b
0 4
(N
s
)

u
b
q
(1)
.
.
.
q
(N
s
)

; (9)
where
T
(s)
=K
(s)
ib
K
(s)
1
ii
(10)
are the static response modes and where 4
(s)
are n
(s)
i
n
(s)

matrices which columns contain the


rst n
(s)

free vibration modes of the substructure clamped on its interface. Applying the reduction
procedure (9), the standard CraigBampton reduced matrices

K
CB
=T
T
CB
K
a
T
CB
and

M
CB
=T
T
CB
M
a
T
CB
are found [2].
D.J. Rixen / Journal of Computational and Applied Mathematics 168 (2004) 383391 387
3.2. Free interface modes as reduction basis
Let us now consider the dual assembled problem (5) or (7). In that case, every substructure can
be seen as being excited through interface connection forces. The local dynamical behavior can
therefore be expressed in terms of eigenmodes of the entire local matrices, hence in terms of free
vibration modes of the substructures with free interfaces, and in terms of local static solutions
u
(s)
=u
(s)
stat
+
n
(s)
m
(s)

r=1
V
(s)
r

(s)
r
; (11)
where
u
(s)
stat
=K
(s)
+
B
(s)
T
z +
m
(s)

i=1
R
(s)
i

(s)
i
: (12)
K
(s)
+
is the inverse of K
(s)
when there are enough boundary conditions to prevent the substructure
from oating when its interface with neighboring domains is free. If a substructure is oating, K
(s)
+
is a generalized inverse of K
(s)
and R
(s)
is the matrix having as column the m
(s)
corresponding rigid
body modes.
(s)
i
are amplitudes of the local rigid body modes and
(s)
r
are the amplitude of the
local eigenmodes.
An approximation is then constructed by considering only the rst n
(s)

free modes V
(s)
in the
expansion. Calling O
(s)
the matrix containing these n
(s)

modes,
u
(s)
K
(s)
+
B
(s)
T
z +R
(s)
z
(s)
+O
(s)
q
(s)
: (13)
Let us observe that, in this approximation, part of the solution in the subspace of O
(s)
is also included
in K
(s)
+
B
(s)
T
z since the generalized inverse has as spectral expansion [4,6]
K
(s)
+
=
n
(s)
m
(s)

r=1
V
(s)
r
V
(s)
T
r
!
(s)
2
r
: (14)
Hence, the approximation (13) can be equivalently written as
u
(s)
=G
(s)
res
B
(s)
T
z +R
(s)
z
(s)
+O
(s)
q
(s)
; (15)
where
G
(s)
res
=
n
(s)
m
(s)

r=n
(s)

+1
V
(s)
r
V
(s)
T
r
!
(s)
2
r
=K
(s)
+

n
(s)

r=1
V
(s)
r
V
(s)
T
r
!
(s)
2
r
(16)
which is computed using the second equality in (16). K
(s)
+
is computed from any generalized
inverse K
(s)

by projecting out the rigid body modes [4,6]. G


(s)
res
is commonly called the residual
local exibility matrix and has the properties
G
(s)
res
=G
(s)
T
res
; O
(s)
T
K
(s)
G
(s)
res
= 0;
G
(s)
T
res
K
(s)
G
(s)
res
=G
(s)
res
; O
(s)
T
M
(s)
G
(s)
res
= 0;
R
(s)
T
M
(s)
G
(s)
res
= 0:
(17)
388 D.J. Rixen / Journal of Computational and Applied Mathematics 168 (2004) 383391
In summary, the local degrees of freedom and the Lagrange multipliers (interconnecting forces) can
be approximated by

u
z

=T
dual

z
(1)
q
(1)
.
.
.
z
(N
s
)
q
(N
s
)
z

R
(1)
O
(1)
0 G
(1)
res
B
(1)
T
.
.
.
.
.
.
.
.
.
0 R
(N
s
)
O
(N
s
)
G
(N
s
)
res
B
(N
s
)
T
0 0 I

z
(1)
q
(1)
.
.
.
z
(N
s
)
q
(N
s
)
z

: (18)
3.3. Reduced matrices and dual assembly
Approximation (18) corresponds to the representation used in substructuring methods such as
the MacNeal method and the Rubin method [3,7,9]. In those procedures, the interface forces are
eliminated locally in order to obtain reduced matrices in terms of the substructure interface displace-
ments only. Hence these procedure go back to a primal assembly expression which simplies the
implementation of the procedure in standard nite element codes. However, the reduced matrices so
obtained have a cumbersome expression and are not as sparse as the CraigBampton matrices.
In this work, we will go the fully dual assembly way: since the approximation (18) has been
shown in the previous section to be related to the dual assembly process, we will reduce the dual
assembled form. Using approximation (18) in the dual formulation (7) and assuming that the rigid
and elastic modes are orthonormalized with respect to M
(s)
, one obtains the hybrid system

z
(s)
q
(s)
.
.
.

+

K

z
(s)
q
(s)
.
.
.
z

=T
T
dual
f (19)
with the reduced hybrid matrices

M =T
T
dual

M 0
0 0

T
dual
=

I 0
0 M
res

(20)
D.J. Rixen / Journal of Computational and Applied Mathematics 168 (2004) 383391 389

K =

0 0
0 D
(1)
2

R
(1)
T
B
(1)
T
O
(1)
T
B
(1)
T

.
.
.
.
.
.
0

0 0
0 D
(N
s
)
2

R
(N
s
)
T
B
(N
s
)
T
O
(N
s
)
T
B
(N
s
)
T

[ B
(1)
R
(1)
B
(1)
O
(1)
] [ B
(N
s
)
R
(N
s
)
B
(N
s
)
O
(N
s
)
] F
res

; (21)
where
F
res
=
N
s

s=1
B
(s)
G
(s)
res
B
(s)
T
M
res
=
N
s

s=1
B
(s)
G
(s)
res
M
(s)
G
(s)
res
B
(s)
T
: (22)
Remark. The following remarks on the reduced matrices (20) and(21) are noteworthy:
The reduced matrices (20) and (21) of the Dual CraigBampton method are similar but dual
to the classical CraigBampton matrices. The cost for building those reduced matrices is also
comparable.
Compared to the Mac Neal and the Rubin methods, the procedure described here enforces only a
weak compatibility between the substructures. Indeed, the last equation in (19) is obtained from
[ B
(s)
G
(s)
res
I ]

.
.
.
M
(s)
u
(s)
+K
(s)
u
(s)
=f
(s)
.
.
.
N
s

s=1
B
(s)
u
(s)
= 0

: (23)
This observation can be interpreted as follows: call f
(s)
the residual forces in the substructures
due to the reduction. Call u
(s)
=G
(s)
res
f
(s)
the displacements these residual would create locally
according to the residual exibility. The weak compatibility condition then expresses that an
interface displacement jump equal to the incompatibility of the u
(s)
is permitted.
It is easy to show that when the eigensolutions associated are computed using inverse iteration
schemes, the associated static problem leads to an interface problem equivalent to the dual interface
problem found in the Finite Element Tearing and Interconnecting (FETI) method [5,8]. Hence the
solution techniques developed for the FETI method can be applied in a straightforward manner to
solve the hybrid problems associated to the reduced system (19).
4. Application examples
Let us consider the example problem (Fig. 1) of a frame made of steel beams. It is divided into
5 substructures and clamped at one end. Each cell in the frame has a height of 0:35 m and a width
390 D.J. Rixen / Journal of Computational and Applied Mathematics 168 (2004) 383391
red full
Dual C.-B.
MacNeal
Craig-Bampton
4 modes/substr.
eigenfrequency number
SUBSTRUCTURE 1
SUBSTRUCTURE 4
SUBSTRUCTURE 3
SUBSTRUCTURE 5
SUBSTRUCTURE 2
1 8 7 6 5 4 3 2
10
0
10
-5
10
-4
10
-3
10
-2
10
-1
10
-6
2

_
full
2

Fig. 1. CraigBampton and dual CraigBampton reduction of a beam frame.


and depth of 0:5 m. All outer beams have a hollow circular cross-section, the outside and inside
diameters being 0.02 and 0:018 m. The diagonal members inside the cells have plain circular section
of diameter 0:008 m.
In Fig. 1 we plot the relative error on the rst eigenfrequencies computed by the CraigBampton,
the MacNeal and the Dual CraigBampton methods, compared to the frequencies obtained for the
non-reduced system. We used four non-rigid modes for the substructures of the frame. The results
show that whereas the CraigBampton and the MacNeal methods yield similar accuracy, the Dual
CraigBampton reduction technique leads to an accuracy nearly two orders of magnitude better in
the low frequency range. The MacNeal and the Dual CraigBampton approaches have the same
reduction basis. However, strong interface compatibility is enforced in the MacNeal approach while
the global eigenmodes computed with the Dual CraigBampton satisfy only a weak form of the
interface compatibility. One can thus speculate that the Dual CraigBampton method is remarkably
ecient because it enforces interface compatibility consistent with the reduction basis.
5. Conclusions
We propose a novel substructuring technique based on free interface modes and dual assembly.
Unlike other well-known substructuring techniques using free interface modes, only a weak compat-
ibility is enforced between substructures. This method exhibits strong similarities with the classical
CraigBampton method and can be considered as its dual. The reduced matrices associated to the
Dual CraigBampton method presented here are very similar to the CraigBampton matrices and less
cumbersome then reduced matrices obtained by other mode synthesis procedures based on free inter-
face modes. Numerical results indicate that the Dual CraigBampton exhibits remarkable eciency.
References
[1] R.R. Craig, Coupling of substructures for dynamic analyses: an overview, in: Structures, Structural Dynamics and
Material Conference, 41st AIAA/ASME/ASCE/AHS/ASC, Atlanta, 2000, AIAA-2000-1573.
[2] R. Craig, M. Bampton, Coupling of substructures for dynamic analysis, Amer. Inst. Aero. Astro. J. 6 (7) (1968)
13131319.
D.J. Rixen / Journal of Computational and Applied Mathematics 168 (2004) 383391 391
[3] R.R. Craig, C.J. Chang, On the use of attachment modes in substructure coupling for dynamics analysis, in: Proceedings
of the 18th Structures, Structural Dynamics and Material Conference, AIAA/ASME, San Diego, AIAA 77-405, 1977,
pp. 8999.
[4] C. Farhat, D. Rixen, Encyclopedia of Vibration, Academic Press, New York, 2002, pp. 710720, ISBN 0-12-227085-1
(Chapter Linear Algebra).
[5] C. Farhat, F.X. Roux, Implicit parallel processing in structural mechanics, Comput. Mech. Adv. 2 (1) (1994) 1124.
[6] M. G eradin, D. Rixen, Mechanical vibrations. Theory and Application to Structural Dynamics, 2nd Edition, Wiley,
Chichester, 1997.
[7] R. MacNeal, A hybrid method of component mode synthesis, Comput. Structures 1 (4) (1971) 581601.
[8] D. Rixen, Encyclopedia of Vibration, Academic Press, New York, 2002, pp. 9901001, ISBN 0-12-227085-1 (Chapter
Parallel Computation).
[9] S. Rubin, Improved component-mode representation for structural dynamic analysis, Amer. Inst. Aero. Astro. J. 13
(8) (1975) 9951006.

You might also like