FACTA UNIVERSITATIS

Series: Mechanical Engineering Vol. 12, No 1, 2014, pp. 37 - 50

Original scientific paper

APPLICATION OF THE CRAIG-BAMPTON MODEL ORDER

REDUCTION METHOD TO A COMPOSITE STRUCTURE:
MAC AND XOR
UDC (531+624.01)

Humberto Peredo Fuentes, Manfred Zehn

Institute of Mechanics, Technical University Berlin, Germany
Abstract The Craig-Bampton model order reduction (CBMOR) method based on the
Rayleigh-Ritz approach is applied to dynamic behavior simulation of a composite
structure in order to verify the method's feasibility and accuracy. The principle of this
method is to represent a coupled component model based on the mass, damping and
stiffness matrices. The methodology consists of a finite element model based on the
classical laminate theory (CLT), a design of experiment to improve the modal
assurance criteria (MAC) and experimental results in order to validate the reduced
model based on CBMOR method and substructures (super-elements). Experimental
modal analysis has been performed using a scanner laser Doppler vibrometer (SLDV)
in order to assess the quality of the finite element models. The MAC and cross
orthogonality MAC (XOR) values are computed to verify the eigenfrequencies and
eigenvectors. This approach demonstrates the feasibility of using CBMOR for
composite structures. The example is prepared and solved with MSC/NASTRAN
SOL103. The design of experiments (DOE) method has been applied in order to
identify the critical parameters and thus obtain high MAC values.
Key Words: SDTools-MATLAB, NASTRAN, Modal Analysis, Composites

1. INTRODUCTION
Many techniques have been proposed to obtain reduced order finite element models
(known as model order reduction (MOR) methods) by reducing the order of mass and
stiffness matrices of structures made of conventional materials [1-3]. The substitution of
conventional materials by composite materials in the aeronautic, space and automotive
industry is becoming increasingly important today for the production of industrial highperformance components [11-13]. The state-of-the-art MOR techniques are classified in
Received November 27, 2013 / Accepted February 4, 2014
Corresponding author: Humberto Peredo Fuentes
Technical University Berlin, Institute of Mechanics, Berlin, Germany
E-mail: hperedo@mailbox.tu-berlin.de

38

H. PEREDO FUENTES, M. ZEHN

four groups [19]: direct reduction, modal methods, reduction with Ritz vectors and the
component mode synthesis (CMS). According to this classification, the last two groups
yield the best results. The Ritz vectors improve the accuracy-cost ratio and the CMS
combines the first three classes of methods. Hence the MOR method based on the
Rayleigh-Ritz approach is used to improve the accuracy-time ratio in civil and
aeronautical engineering applications in many areas of structural dynamics [6, 14, 19, 22,
23]. Thus, it is necessary to study the feasibility and efficiency of using the CMS with the
Rayleigh-Ritz reduction basis in order to describe the dynamic behavior of a composite
structure [14, 19]. The sections 2-4 introduce to MOR based on the Ritz vectors, classical
CMS and substructures, respectively. The classical laminate theory (CLT) is introduced in
Section 5. Sections 6-8 demonstrate a sensitivity analysis performed by using different
tools design of experiment (DOE), finite element method (FEM) and modal assurance
criteria (MAC).

2. MODEL ORDER REDUCTION WITH RITZ VECTORS

It is typical for coupled problems with model sub-structuring [6, 14, 22, 23] to have an
accurate second order representation in the form:
([ M ]s 2 [C ]s [ K ]){q} [b]{u}
{ y} [c]{q}

(1)

where s is the Laplace variable, [M], [C], [K] are mass, damping and stiffness matrices,
respectively, {q} are generalized degrees of freedom (DOFs), [b] and [c] are input and
output matrices, respectively, {u} are the inputs describing the time/frequency dependence,
and {y} are the physical outputs.
In this description, two - not very classical and yet important - assumptions are made:
1) The decomposition of discretized loads F(s) as the product of the fixed input shape
matrix specifying the spatial localization of loads [b] and inputs {u}.
2) The definition of physical outputs {y} is a linear combination of DOFs {q}.
The Ritz/Galerkin displacement methods seek approximations of the response within a
subspace characterized by matrix [T] associated with generalized DOFs {qR}:
{q} [T ]{q R } ,

(2)

where {q} is the original set of DOF and {qR} is the reduced set of DOF, substituting Eq.
(2) into Eq. (1) leading to an overdetermined set of equations. The Ritz approximation
assumes that the virtual work of displacements in the dual subspace generated by [T] T is
also zero, thus leading to a reduced model:

[T ] [M ][T ]
T

s 2 [T ]T [C ] [T ] s [T ]T [ K ] [T ] {qR (s)} [T ]T [ B] {u (s )}
{ y( s)} [c] [T ] {qR ( s)}

(3)

Application of the Craig-Bampton Model Order Reductio Method to a Composite Structure: MAC and XOR 39

3. CLASSICAL CMS BASES AS APPROXIMATION OF THE FREQUENCY RESPONSE

The method was first developed by Walter Hurty in 1964 [1] and later expanded by Roy
Craig and Mervyn Bampton [2] in 1968. Component mode synthesis and model reduction
methods provide for the means for building appropriate [T] bases (the subspace spanned
rectangular matrix). There are many ways of proving classical bases [22]. Their validity is
associated with two assumptions: the model needs to be valid over a restricted frequency
band and the number of inputs is limited. One needs to translate this hypothesis into the
requirement to include mode shapes and static responses into [T] basis. Most of the literature
on CMS implies the fundamental assumption for coupling, which states that the
displacement is continuous at the interfaces. Considering the response of an elastic structure
to applied loads F(s)=[b]{u(s)}, the exact response at a given frequency [H(s)] is given by:

[ H (s)] [c]{[M ]s 2 [ K ]}1[b] [c][Z (s)]1[b] ,

(4)

where [Z(s)] is the dynamic stiffness. If there is no external excitation:

[Z ( j )]1{ j } {0} ,

(5)

and the solutions are known as free modes of the structure, where j is jth eigenvalue of
the matrix and {j} is jth eigenvector. A reduction model should include these shapes to
allow for an accurate representation of the resonances which are associated with the
singularities of the dynamic stiffness. A point of particular interest is the static response at
s=0. The associated deformation is:

{q(s 0)} [Z (0)]1[b]{u(0)} [Ts ]{u(0)} .

(6)

The columns of [Ts] are also called attachment modes [22]. For the case of free
floating structures (structures with rigid modes), [Z(0)] is singular and one defines
attachment modes as responses of all modes except for the rigid body modes.
The bases combining free modes and attachment modes are valid over a certain frequency
range (truncation of the series of free modes) and certain inputs characterized by [b].
One, thus, considers the response of the structure with enforced displacements on a
subset of DOFs. Division of the DOFs in two groups active or interface DOFs denoted
by I in the subscript, and complementary, denoted by C in the subscript, leads to:

[ Z II ( s)] [ Z IC ( s)] {qI ( s)} RI ( s)

[ Z ( s)] [ Z ( s)] q ( s) {0} ,

CC
C
CI

(7)

where <{qI(s)}> and <{0}> denotes a known quantity. The exact solution to this problem is:

[I ]

{q} [T ( s)]{qI }
{qI } .
1

[
Z
(
s
)]
[
Z
(
s
)]
CC
CI

(8)

The subspace found here is frequency dependent and can only be used in very
restricted applications [23]. A classical approximation is to evaluate the static (s=0) value
in this subspace for the active or interface DOFs denoted by CI in the subscript, and
complementary, CC in the subscript:

40

H. PEREDO FUENTES, M. ZEHN

[I ]

[T ]
.
1
[ KCC ] [ KCI ]

(9)

Reduction on this basis is known as static or Guyan condensation [4]. The columns of
[T] are called constraint modes [22]. They correspond to unit displacements of the
interface DOFs. Significant deviations can be expected when [ZCC(s)]-1 differs from
[ZCC(0)]-1=[KCC]-1 Such difference is significant for singularities of [ZCC(s)]-1 which are
computed by the eigenvalue problem:

[0] 0
[0]
[0] [ Z ( )] 0 .
CC
j j ,c

(10)

The use of a basis combining constraint, Eq. (9), and fixed-interface modes, Eq. (10),
is proposed in [2]. It yields the Craig-Bampton method:

[I ]
[0]

[T ]
,
1
[ KCC ] [ KCI ] [NM,C ]

(11)

where [NM,C] is the interior part of the matrix of kept fixed-interface modes. There are many
results reported by Balms et al. [6, 14, 15] obtained by the Craig-Bampton model order
reduction (CBMOR) and the Rayleigh-Ritz vectors approach in order to solve coupled
problems related to model sub-structuring (also known as component mode synthesis).
One should be aware of the fact that the use of Raleigh-Ritz vectors leads to dense matrices, as
opposed to not reduced FEM models characterized by a sparse form of the matrices.

4. SUBSTRUCTURES OR SUPER-ELEMENTS
Sub-structuring is a procedure that condenses a group of finite elements into one
element. It implies that the whole structure is divided into smaller structures (see Figs. 1
and 2), and the resulting elements are referred to as super-elements. In the considered
case (Fig. 1), the structure is divided into two substructures using 123 nodes at the
interface. The model size is reduced from 37,698 DOF to 579 DOF.

a)

b)

Fig. 1 Prototype and FEM model in NASTRAN and SDTools:

a) Composite structure front and back; b) FEM model

Application of the Craig-Bampton Model Order Reductio Method to a Composite Structure: MAC and XOR 41

The basic sub-structuring idea is to consider a part of the model separately and extract
the degrees of freedom needed to connect this part to the rest of the model. Therefore, the
result of sub-structuring is a collection of finite elements whose response is defined by the
stiffness and mass of the retained degrees of freedom. The categories of modal truncation
sub-structuring and static condensation approaches have been widely applied relying on the
eigenfrequency information [3, 23].

a)

b)

Fig. 2 Sub-structuring: a) Substructure 1; b) Substructure 2

5. LAMINATE THEORY
The classical laminate theory is applicable to linear and composite elastic materials [21]
by means of the Discrete Kirchhoff Theory (DKT) elements [20]. The CLT has been used
extensively to predict elastic behavior of the traditional fiber-reinforced polymers (FRP).
FRP materials (carbon or glass FRP) are widely used in aerospace and construction
applications. One important consideration is to have perfectly bonded layers with a uniform
thickness (see Fig. 3). The mechanical properties measured in ply level experiments are used
to populate the stiffness matrix for each ply. The stiffness matrices for the individual plies
are combined to form the laminate stiffness matrix the ABC matrix:

x
Nx

y
Ny
N xy A B xy
.

M x B C x

M
y
y

M
z
xy

(6)

The ABC matrix relates forces (Ni) and moments (Mi) to strains (i) and curvatures
(i). The components of the ABC matrix are given in Eqs. (7-9), where N is the number of
plies, Qk is the stiffness matrix of each ply, and Zk denotes the distance from the laminate's
mid-plane to the edges of single plies:
N

A Qk [ Z k Z k 1 ], in-plane stiffness matrix,

k 1

(7)

42

H. PEREDO FUENTES, M. ZEHN

1 N
Qk [Zk2 Zk21 ], bending-stretching coupling matrix,
2 k 1

(8)

1 N
Qk [Zk3 Zk31 ], bending-stiffness matrix.
3 k 1

(9)

Fig. 3 Configuration of composite layers

The prototype and the finite element (FE) model are shown in Fig. 1. The composite
structure incorporates three parts (properties in Table 1). The first component is made of
Hunts-man Ly 564 + Hexcel Gewebe G0926 (HTA-Faser) with dimensions of 0.390m
0.810m (Fig. 2). The middle shell that connects the two principal parts (Fig. 2a) has
dimensions of 0.710m 0.030m. Finally, there is the C-section Hexcel RTM6 + Saertex
Multi-Axial-Gelege (MAG) with a IM7-Faser with dimensions of 0.710m 0.030m. All
the parts have symmetric layer distribution [45/-45/45/-45/]S.

6. DESIGN VIA FINITE ELEMENT ANALYSIS (FULL AND REDUCED MODEL)

Our study is divided into two parts.
The first part is a full modal analysis using the same model but with two different
solvers for reference purposes. Two types of elements have been used: CTRIA3 shell
(from MSC/NASTRAN) and PSHELL (from SDTools).
The second part is setting the reduced model by using SDTools for MATLAB. The
reduced model is built up defining two super-elements. Super-element 1 (Fig. 2a) has 4,753
nodes and 9,219 elements, while super-element 2, (Fig. 2b) has 1,615 nodes and 3,026
elements. The defined super-elements share 579 DOF distributed in 123 nodes along the
common border with different DOF per node, according to the CMS that has defined an
appropriate [T] matrix, used in [3]. We have calculated the same number of modes in each
super-element and performed a cross orthogonality MAC (XOR) evaluation to verify the
approximation of the MOR used in low (12 mode pairs) and/or high frequency range (29
mode pairs) versus the full model.
In order to estimate the main parameters (qualitative and quantitative) that affect our
MOR based on the number of substructures and nodes, we have performed a DOE using
first the full model and the experimental analysis. The DOE study is performed using the
methodology implemented in Minitab 16 [7]. Fig. 4a shows the main effects of each
parameter in the composite structure based on the physical characteristics selected. The

Application of the Craig-Bampton Model Order Reductio Method to a Composite Structure: MAC and XOR 43

main effect is identified through the slope generated due to the eigenfrequency values
between the limits defined for each parameter a bigger slope means a strong parameter
effect. Due to the number of parameters, it is necessary to perform first a DOE-screening
with 210-5=32 "runs" and then a full factorial with the identified principal parameters
based on the DOE-screening.

a)

b)

Fig. 4 DOE: a) Parameters main effects, b) Surface response

The results shown in Fig. 5a (vertical left side) are eigenfrequencies. The MAC
correlation between the full model and experimental data help us validate the MOR
results. The Young Modulus, density, number of nodes and substructure parameters have
a strong influence reflected in the slope (Fig. 4b) and in the MAC values (section 7).
Once we have selected the main parameters based on the DOE-screening, we perform a
DOE full factorial 24 and obtain a surface response (see Fig. 4b) that help us find the best
model for the parameter limits selected. This process is known in literature as updating.
Jing [8], Barner [9] and Xiaoping et al. [10] reported the use of design of experiments in
order to quantify and qualify different key parameters in mechanical components
(stresses, displacements, low and high cycle fatigue, and frequencies). The DOE is a
sensitivity analysis tool used to estimate the critical input parameters.

a)

b)

Fig. 5 Cross orthogonality MAC reduced vs. full model: (a) low frequencies
(b) higher frequencies (green bars MAC, blue bars frequency difference)

44

H. PEREDO FUENTES, M. ZEHN

In Fig. 5, we can see the low and high mode pairs selected between the full and
reduced model (12 and 29 mode pairs), respectively. The green bars show the eigenvector
criteria and the blue bars the eigenfrequency difference between the reduced and the full
model. The low frequencies show a larger difference in the 3rd, 4th and 12th mode pair. The
largest difference in the frequencies is about 1.2% (low eigenfrequencies) between the full
and reduced model. Increasing the number of pairs, the eigenfrequency difference increases
up to 3% for 29 pairs. However, the mode pairs 3, 4 and 12 have improved suggesting that
the accuracy using CBMOR method depends on the number of retained constraint modes.
Most of the pair selections have a correlation above 90%, except for the 12th mode pair in
the low frequency range and the 23rd and 24th pair in a high frequency range. Table 2 shows
the values comparing the full with a reduced model for low frequency. A 3D plot of the
XOR for high frequency pairs is given in Fig. 6.
Table 1 Orthotropic elastic mechanical properties per thickness
Modulus
E1
E2

Th1(m)
0.035

E(GPa)
71.3
97.3

(-)
0.3
0.3

Modulus
E1
E2

Th2(m)
0.007

E(GPa)
71.3
68.3

(-)
0.2
0.2

Modulus
E1
E2

Th3(m)
0.035

E(GPa)
71.3
68.3

(-)
0.2
0.2

Shear
G1
G2
G3
Shear
G1
G2
G3
Shear
G1
G2
G3

G(GPa)
7.0
5.0
7.0
G(GPa)
6.0
5.0
6.0
G(GPa)
6.0
5.0
6.0

(Kgm-3)
2600

(Kgm-3)
1500

(Kgm-3)
1500

Table 2 MAC values: full versus CBROM reduced model

7
8
9
10
11
12
13
14
15
16
17
18

Full
57.218
106.02
167.50
168.2
234.99
236.83
315.26
323.93
401.72
408.39
432.89
494.90

7
8
9
10
11
12
13
14
15
16
17
18

Reduced
57.218
106.21
167.79
168.29
235.12
236.93
315.33
326.82
401.77
408.57
433.39
501.41

DF/FA
0.0
0.2
0.2
0.1
0.1
0.0
0.0
0.9
0.0
0.0
0.1
1.3

MAC
100
100
93
94
100
100
100
98
100
100
99
69

The correlation of nearly double modes 9-10,11-12,13-14 and 15-16 in Table 2

suggests the possibility of having bending and torsional modes at close frequencies in the
composite structure (mode veering) [24]. Thus, a lower MAC value is expected in some
mode pairs in the experimental validation. There are only three types of structures made of

Application of the Craig-Bampton Model Order Reductio Method to a Composite Structure: MAC and XOR 45

the conventional materials that have been identified to exhibit veering: symmetric or
cyclic structures, multi-dimensional structures such as plates having bending and torsion at
close frequencies and structures with fully uncoupled substructures. The considered
structure corresponds to the second type multi-dimensional plate structures.

7. MODAL ASSURANCE CRITERION (MAC)

There are two general categories for correlation criteria: eigenfrequencies and
eigenvectors [18]. The MAC is one of the most useful comparison methods that relies on the
eigenvector information according to Eq. (10). The MAC is a known vector correlation
between the experimental and the FE model. To approximate the measurements through a
polynomial function, (Fig. 9), we use the frequency domain identification of structural
dynamics applying the pole/residue parameterization [15].

MAC (i )

(c j id ) H (c j k )
j 1

(c j id )
j 1

(10)

(c j id ) (c j k ) (c j k )
H

j 1

The MAC value of 100 % corresponds to an absolute correlation. The less this value
becomes, the worse the eigenvector correlation is (cjid is the jth mode shape at sensors
and cjk is the jth analytical mode shape), provided that the observability law for the
selection of DOFs is not violated. A MAC coefficient of a magnitude larger or equal than 90%
implies a satisfactory correlation. In Fig. 8, we observe some mode shapes of the reduced and full
models. Figs. 10a, 10b, and 10c, show the MAC between the full and the experimental
measurements in MATLAB, NASTRAN, and CBMOR model, respectively. The correlation
is performed for a low frequency range (up to 400 Hz), based on the fitting model generated
from the experimental measurements [3, 15].

Fig. 6 Cross orthogonality MAC (XOR): higher frequencies reduced vs. full model

46

H. PEREDO FUENTES, M. ZEHN

Fig. 7 FRF(blue) and fitting curve (green) of composite model at node 183y

8. EXPERIMENTAL MODAL ANALYSIS

All the measurements are performed with the Scanning Laser Doppler Vibrometer
(SLDV) PSV 840 (Fig. 9a). It is a complete and compact system including a sensor head,
a PC with DSP boards and Windows NT-based application software packages [16].
Discrete Fourier transform is applied to response x(t) and excitation f(t) to give X(i) and
F(i), respectively [17]. The frequency response function (FRF), H(i), is defined as the
ratio of the transformed excitation [18]:

H (i )

X (i )
,
F (i )

(11)

where H(i) is the identified (predicted) FRF transfer function matrix , H(i) the measured
FRF transfer function matrix, X(i) the Fourier spectrum of response, and F(i) is the
Fourier spectrum of excitation force. The FRF in Eq. (11) is the inverse of the dynamic
stiffness matrix:

H (i ) [i2 [M ] i [C] [ K ]]1 .

(12)

Mass [M], damping [C] and stiffness [K] matrices in Eq. (12) are dependent on physical
parameters such as material's density, Young's and shear moduli and Poisson ratio.
Mode 7 at 57.22 Hz,
Mode 7 at 57.22 Hz reduced

Mode 8 at 106 Hz,

Mode 8 at 106.22 Hz reduced

Mode 9 at 167.5 Hz,

Mode 9 at 167.8 Hz reduced

Mode 10 at 168.2 Hz,

Mode 10 at 168.3 Hz reduced

Fig. 8 CBMOR (in green) vs full model in MATLAB (in blue)

Application of the Craig-Bampton Model Order Reductio Method to a Composite Structure: MAC and XOR 47

a)

b)

c)

Fig. 9 Experiment: a) Experimental set-up; b) 153 Y-direction sensors; c) 153 sensors in

the FEM model
The SLDV employs a laser to sweep over the structure continuously while measuring,
capturing the response of the structure from a moving measurement point. Various methods
have been devised to determine the mode shapes of the structure everywhere along the scan
path measurement [16]. A bandwidth of 2% is used in order to localize the eigenfrequencies.
The composite structure has rather small internal damping and the experimental modal
analysis below 400 Hz is performed. The structure is excited by means of a shaker at node
17 (Fig. 9a and Fig. 9b) that is located in the right bottom corner. The input force is
measured using a force transducer type 8200 in combination with a charge to CCLD
converter Type 2646 in order to record the excitation in the transverse direction.
The interpolation between the experimental measurements uses Frequency Response
Functions (FRF) [15], (Fig. 7). The FRFs allow comparison of the experimental modal
parameters (frequency, damping, and mode shape) with those of the FE model. The Fast Fourier
Transform (FFT) is a fundamental procedure that isolates the inherent dynamic properties of a
mechanical structure and in our case with respect to the full and reduced FE model. The MAC
analysis (Fig.10) shows a high correlation between the full model, the reduced model and
the experimental measurements. The nearly double correlation in the experimental results
identified in Table 3 (previously identified applying the CBMOR method in Table 2),
suggests the presence of the veering phenomena (bending and torsional mode at the same
frequency) in the considered composite structure. This is reflected in the MAC values for
the corresponding modes.

a)

b)

c)

Fig. 10 Comparative MAC: a) SDtools-Exp, b) MSC/NASTRAN-Exp, c) CBMOR-Exp

48

H. PEREDO FUENTES, M. ZEHN

Fig. 11 Experimental mode shapes

Pierre [25] reported how localization and veering are related to two kinds of "coupling":
the physical coupling between structural components, and the modal coupling set up
between mode shapes through parameter perturbations.
His studies show that, in structures with close eigenvalues, small structural irregularities
(could be our case) result in both strong localization of modes and abrupt veering away of
the loci of the eigenvalues when these are plotted against a parameter representing the
system disorder. The study of the presence of this phenomenon in the composite structure is
beyond the scope of this work.
Table 3 shows the MAC values obtained for each case between the full and reduced
model versus the experimental results. The mode shapes depicted in Fig. 11 are the
experimental results.
Table 3 Full and reduced FEM model results versus experimental results
Experimental
1
49.243
2
92.265
93.756
3
4
145.29
160.05
5
164.18
6
226.36
7
243.40
8
307.33
9
10
314.18
11
324.83
12
329.67

7
8
8
10
10
9
12
11
14
14
13
13

Full
57.218
106.02
106.02
168.20
168.20
167.50
236.83
234.99
323.93
323.93
315.26
315.26

DF/FA
16.2
14.9
13.1
15.8
5.1
2.0
4.6
-3.5
5.4
3.1
-2.9
-4.4

MAC
100
97
90
83
86
98
86
96
81
66
74
90

CBMOR
57.218
106.21
106.21
168.29
168.29
167.79
236.93
235.12
326.82
326.82
315.33
315.33

DF/FA
16.2
15.1
13.3
15.8
5.1
2.2
4.7
-3.4
5.4
4.0
-2.9
-4.3

MAC
100
97
90
84
71
92
85
97
80
65
74
89

Application of the Craig-Bampton Model Order Reductio Method to a Composite Structure: MAC and XOR 49

9. CONCLUSIONS
The results have shown a good correlation in dynamic behavior of the composite
structure model using the DKT elements with different solvers. The MAC values with the
full and reduced models have also shown a good agreement with the experimental results. In
order to achieve high quality models that can adequately capture the dynamic behavior, the
material properties are updated through the DOE and are crucial in the MOR correlation
with the experimental results. The updated mass and stiffness matrices in the full model play
an important roll in this procedure. Furthermore, the reduced model obtained by means of
the Craig-Bampton MOR method (the reduced model couples 2 substructures through 123
nodes and 579 DOF) has demonstrated a good agreement with the experimental results. The
MAC values for the FEM models as well with the experimental results suggest a presence of
mode veering phenomenon (bending and torsional mode at the same frequency in the
considered composite structure). And finally, the experimental results using a SLDV as well
as the identification of pole/residues used in [15], are suitable to validate the dynamic
analysis using modal order reduction. It is improper to draw conclusions from a single
example, but the obtained results using two different solvers are coherent. This conducted
work obviously leaves much room for further research. Other modal assurance criteria need
to be performed, such as coordinate modal assurance criteria (COMAC), enhanced modal
assurance criteria (ECOMAC) and scale coordinate assurance criteria (S-COMAC) and also
other model order reduction and/or mode shape expansion methods should be assessed.

REFERENCES
1. Hurty, W. C., 1965, Dynamic analysis of structural systems using component modes, AIAA Journal,
3(4), pp. 678-685.
2. Craig R. J. and Bampton M., 1968, Coupling of substructures for dynamic analyses, AIAA Journal 6(7),
pp.1313-1319.
3. SDTools Inc. 2011, Structural dynamics toolbox and FEMLink, User's Guide, SDTools, Ver. 6.4, Paris,
France.
4. Guyan, J. 1965, Reduction of stiffness and mass matrices, AIAA Journal, 3(380). pp.
5. Irons, B. M.., 1965, Structural eigenvalue problems - elimination of unwanted variables, AIAA Journal,
3(5): pp. 961-962.
6. Balms E., 1996, Use of generalized interface degrees of freedom in component mode synthesis,
International Modal Analysis Conference, pp. 204-210.
7. Montgomery, D. C., 2000, Design and analysis of experiments, John Wiley & Sons.
8. Fan J., Zeng, W., Wang R., Sherr X., Chen Z., 2010, Research on design and optimization of the
turbine blade shroud, 2nd International Conference on Engineering Optimization, Lisbon, Portugal.
9. Barner, N., 2010, Isight-Abaqus optimization of a ring-stiffened Cylinder, SIMULIA Customer Conference.
10. Chen, X., Yu, X., and Ji B., 2010, Study of crankshaft strength based on iSIGHT platform and DOE
methods, International Conference on Measuring Technology and Mechatronics Automation, pp. 548-551.
11. Lauwagie, T., 2005, Vibration-Based Methods for the Identification of the Elastic Properties of Layered
Materials, PhD thesis, Catholic University of Leuven, Belgium.
12. Reddy, J. N., 2005, Mechanics of Laminated Composite Plates and Shells Theory and Analysis, CRC,
Press Second edition.
13. Berthelot, J. M., 1992, Materiaux composites: Comportement mecanique et analyse des structures,
Lavoisier, Paris, France
14. Balms E., 1997, Efficient Sensitivity Analysis Based on Finite Element Model Reduction, International
Modal Analysis Conference, IMAC, pp.1-7.
15. Balms E., 1996, Frequency domain identification of structural dynamics using the pole/residue
parametrization, International Modal Analysis Conference, pp. 540-546.

50

H. PEREDO FUENTES, M. ZEHN

16. Gade, S., Mller, N.B., Jacobsen, N.J., and Hardonk. B., 2000, Modal analysis using a scanning laser Doppler
vibrometer, Sound and Vibration Measurements, pp. 1015-1019.
17. Newland, D.E., 1993, An Introduction to random vibration, spectral and wavelet Analysis, New York, Longman,
Harlow and John Wiley.
18. Ewings, D. J., 1995, Modal testing: Theory and practice, Research Studies Press, Letchworth, United Kingdom.
19. Cunediolu, Y., Muan, A., Akay, H., 2006, Frequency domain analysis of model order reduction techniques,
Finite Elements in Analzsis and Design, 42, pp. 367-403.
20. Batoz, J.L., Bathe, K.J., Ho, L.W., 1980, A Study of three node triangular plate bending elements, International
Journal for Numerical Methods in Engineering, 15, pp. 1771-1812.
21. Batoz, J.L., Lardeur, P., 1989, Composite plate analysis using a new discrete shear triangular finite element,
International Journal for Numerical Methods in Engineering, 27, pp. 343-359.
22. Craig, R.J., 1987, A review of time-domain and frequency domain component mode synthesis methods. Int. J.
Anal. and Exp. Modal Analysis, 2(2), pp. 59-72.
23. Balms, E., 2000, Review and Evaluation of shape expansion methods, International Modal Analysis Conference,
pp. 555-561.
24. Bonisoli E, Delprete C., Espoito M., Mottershead J. E., 2011, Structural Dynamics with conicident Eigenvalues:
Modeling and Testing, Modal Analysis Topics 3, pp 325-337.
25. Pierre C., 1988, Mode Localization and eigenvalue loci of Bridges with Aeroeslastic effects, Journal of
Engineering Mechanics 126(3), pp. 485-502.

PRIMENA CRAIG-BAMPTON REDUKCIJE MODELA NA

STRUKTURU OD KOMPOZITNOG MATERIJALA: MAC I XOR
Craig-Bampton metoda za redukciju modela (CBMOR) zasnovana na Rayleigh-Ritz pristupu je
primenjena u simulaciji dinamikog ponaanja kompozitnih struktura u cilju verifikacije izvodljivosti i
tanosti ove metode. Princip ove metode je da predstavi model spregnutih komponenti preko matrica
inercije, priguenja i krutosti. Metodologija ukljuuje model primenom konanih elemenata (MKE) na
osnovu klasine teorije laminata (CLT), zatim postavku eksperimenta sa ciljem poboljanja vrednosti
koeficijenata poreenja modova (MAC), kao i eksperimentalne rezultate sa ciljem validacije
redukovanog modela primenom CBMOR metode i substruktura (superelemenata). Eksperimentalna
modalna analiza je sprovedena korienjem laserskog Doplerovog vibrometra da bi se ocenio kvalitet
MKE modela. MAC vrednosti za pripadajue i nepripadajue modove su sraunate da bi se
verifikovale sopstvene frekvence i modovi. Ovaj postupak pokazuje izvodljivost primene CBMOR
redukcije modela u sluaju kompozitnih struktura. Model je pripremljen i reen primenom
programskog paketa MSC/NASTRAN SOL103. Metodom dizajna eksperimenta identifikovani su
kritini parametri, to je kasnije omoguilo dobijanje visokih MAC vrednosti.
Kljune rei: SDTools-MATLAB, NASTRAN, modalna analiza, kompozitni materijali