ARTICLE IN PRESS

Mechanical Systems
and
Signal Processing
Mechanical Systems and Signal Processing 19 (2005) 123–138
www.elsevier.com/locate/jnlabr/ymssp

Improvement of damage-assessment results using high-spatial

density measurements
R. Pascuala,*, R. Sch.alchlib, M. Razetob
a
Department of Mechanical Engineering, Universidad de Chile, Casilla 2777, Santiago, Chile
b
! Casilla 53-C, Concepcion,
Department of Mechanical Engineering, Universidad de Concepcion, ! Chile
Received 19 August 2003; received in revised form 30 January 2004; accepted 11 February 2004

Abstract

Model-based damage assessment is based on measuring the distance between experimental and analytical
results. In practice, measurements yield only partial mode shapes with respect to the total degrees of
freedom present in the corresponding finite element model. Thus, before any damage detection method is
implemented, the experimental mode shape has to be expanded to the same dimension of the numerical
mode shape. Mode shapes expansion is a key point in the damage localisation process, since actual defects
of the structure may be hidden by expansion errors. This paper introduces a new general procedure to the
expansion/damage assessment process using an optimised choice for: the size of the expansion basis, the
number of experimental degrees of freedom and the sensor placement. We introduce a new indicator to
evaluate the problems inherent to the expansion/damage detection process using the minimisation of error
on constitutive equations (MECE) technique. It provides insight of the inherent limitations of MECE and
helps the decision making process on how many degrees of freedom should be measured and how many
mode shapes should be used in the expansion basis. The procedure is illustrated using a finite element model
of a plate-like structure, where the damage state is simulated as a reduction of the local stiffness.
r 2004 Elsevier Ltd. All rights reserved.

Keywords: Mode shape expansion; High-spatial density measurements; Damage detection; Finite element

1. Introduction

In recent years, a great effort has been put forward to assess damage in structures using
vibration measurements. The vibration and model based damage identification methods rely on
the fact that the occurrence of damage in a structural system leads to changes in the

*Corresponding author. Tel.: +56-2-6784591; fax: +56-2-6896057.

E-mail address: [email protected] (R. Pascual).

0888-3270/$ - see front matter r 2004 Elsevier Ltd. All rights reserved.
doi:10.1016/j.ymssp.2004.02.004
 ARTICLE IN PRESS

124 R. Pascual et al. / Mechanical Systems and Signal Processing 19 (2005) 123–138

Nomenclature

K; M finite element model stiffness and mass matrices

K ; M experimental matrices
o experimental frequency
Z dynamic stiffness matrix at o
DZ error in the dynamic stiffness matrix
f ith numerical mode shape
Df residual force vector related to v
%
o measured frequency for v
v expanded experimental mode shape
v experimental vector
v measured mode shape
u instrument mode shape associated to v
Ks sth elementary stiffness matrix in global coordinates
e
K condensed stiffness matrix
es energy in element s
Y equilibrium weight matrix
X regularisation weight matrix
F model modal matrix
q modal displacements
q perfectly estimated modal coordinates
a control parameter
nn number of numerical modes employed for expansion
nfe number of degrees of freedom of the FE model
ne number of elements of the FE model
nm number of measured degrees of freedom
T expansion operator
m modal mass matrix associated to F

Superscripts
 experimental
se SEREP
me MECE

Subscripts
m measured set of degrees of freedom

Abbreviations
FE Finite element
SEREP System equivalent Reduction Expansion Process
MECE Minimisation of Errors in Constitutive Equations
DOFs Degrees Of Freedom
 ARTICLE IN PRESS

R. Pascual et al. / Mechanical Systems and Signal Processing 19 (2005) 123–138 125

dynamic properties of the structure. These changes may be used to assess the state of damage of a
structure.
In this article, we introduce a new indicator to investigate the problems inherent to the
expansion/damage localisation procedure using the minimisation of errors on constitutive
equations (MECE) technique [1]. This indicator allows us to evaluate the limitations that appear
when the expanded mode shapes are a combination of a limited set of FE eigenmodes. A
considerable restriction in the number of analytical mode shapes in the expansion basis may cause
a deficient error localisation, because actual defects of the structure may be hidden by expansion
errors.

1.1. State of the art

Several model based structural damage detection methods can be found in the literature. They
vary on how and what experimental data is used in the methods. Some researchers used changes in
modal parameters to locate structural damage; others prefer the forced responses.
Salawu [2] published a survey on the use of natural frequency changes for damage diagnostics.
The shift in the natural frequencies have some practical limitations to damage localisation. The
low sensitivity of the natural frequency shifts to damage requires very precise measurements or
large levels of damage. The modal frequencies are a global property of the structure and they
generally can not provide spatial information about the structural damage.
Ratcliffe [3] used the Laplacian operator on the mode shapes to locate damage. When the
damage is severe, the results are successful. However, when the damage is incipient, further
processing of the Laplacian output is required. Higher frequency mode shapes were not effective
to locate damage since they tend to show only local displacements.
Pandey et al. [4] employed the change in the mode shapes curvature to detect damage. The
curvatures are obtained using a central difference approximation. In a situation with little
displacements, the curvature approximation becomes very sensitive to the presence of noise. The
method seems to work properly with higher frequency mode shapes. However, as the number of
modes increases, measurements become more expensive to carry on.
Shi et al. [5] suggested the use of the change of the modal strain energy in each element before
and after the occurrence of damage. Unfortunately, elements corresponding to nodal points of the
mode shape show exceptionally large and small values of the modal strain energy, and this fact
may give a wrong indication of the damage location.
Lee and Shin [6] introduced a frequency-domain method of structural damage identification.
This method was formulated from the dynamic stiffness equation of motion. The method does not
seem to be suitable for large FE models because the number of unknown damage magnitudes is
much more larger than the measured degree of freedom (DOFs). Thus, it is necessary to derive
more linear algebraic equations to make the damage identification well posed. In other words, as
the finite elements number increases, more measurements are required.
Ren et al. [7] proposed the residual modal force damage index method to predict the damage
location and severity. Their formulation starts from the error of equilibrium resulting from
substituting the measured modal data into the structure eigenvalue equation. An inconvenience is
that the method presents two types of different variables: forces and moments. In order to
 ARTICLE IN PRESS

126 R. Pascual et al. / Mechanical Systems and Signal Processing 19 (2005) 123–138

compare results, it is necessary to scale these two contributions. Nevertheless, the comparison is
not evident.
Messina et al. [8] proposed an assurance criterion to identify the amount of damage at multiple
sites. The method is based on the sensitivity of the natural frequency of each mode to damage in
each site. If the number of potential damaged sites is large, the computational cost becomes very
high. In a multiple damage situation, it is necessary to consider several of the modes shapes whose
frequencies change significantly, and for each of them, to identify a list of potentially damaged
sites. Any restriction to the set of locations to be searched could end in an inaccurate damage
localisation. In the same field, Kim and Stubbs [9] proposed an improved algorithm to locate and
estimate severity of damage in structures using changes in modal characteristics. They assumed
that all the nodal points are measured; however in practice it is almost impossible to have the
same number of experimental and numerical degrees of freedom.
Gawronski and Sawicki [10] proposed a structural damage localisation procedure using modal
and sensor norms. Unfortunately, their approach involves knowing a priori the location of the
damage, since the structures are more densely meshed near the damage locations so as to better
reflect the stress concentration.
Thyagarajan et al. [11] investigated an optimisation procedure to update the FE model matrices
to the damage state. The limitation of the technique is that it requires a lot of computing which
may restrict it to small models. A large number of design variables would cause a divergence of
the optimisation procedure.
In this general scenario we by-pass the difficulties mentioned in the published papers
taking advantage of the great amount of information provided by numerical models (FE) and
high-spatial density vibration measurements.

2. MECE expansion/error localisation

The dynamic equilibrium equation of the experimental structure corresponding to a particular

mode shape v may be written as

K v ¼ o2 M v ; ð1Þ

where K and M respect the same connectivity pattern as the finite element matrices K and M;
respectively. Without any loss of generality, the damping terms are not considered. This
development is intended to explain how the method works. The assumption of such a structural
model for the experimental structure, considers in an implicit way a linear behaviour. Also, there
exists an instrument vector u that satisfies the equilibrium equation:

% 2 Mv;
Ku ¼ o ð2Þ

where K and M are numerical quantities, o% is the measured frequency and v is the experimental
% Note that superscript  refers to actual experimental responses and system
expanded vector for o:
matrices. They differ of the measured responses (noise is present) and also of the expanded
responses (expansion error appear due to noise and model errors). The correspondence between
 ARTICLE IN PRESS

R. Pascual et al. / Mechanical Systems and Signal Processing 19 (2005) 123–138 127

the experimental structure and the FE model may be established through the following relations:
K ¼ K þ DK ;
M ¼ M þ DM ð3Þ
and the assumed experimental mode shape v (and its frequency o ) can be broken down into the
expanded experimental vector v plus the errors Dv and Do:
v ¼ v þ Dv;
o ¼ o
% þ Do ð4Þ

2.1. Expansion of the experimental information

Expansion techniques consider the transformation of the shape vectors exclusively. The
expanded vector v may be sought by minimising the residue of the equilibrium equation in some
adequate metric Y:
min Df T HDf; ð5Þ
v

where
% 2 MÞ v ¼ Z v;
Df ¼ ðK  o ð6Þ
Complementing the objective function (5), the distance between the expanded vector and the
experimental vector may be also minimised:
min ðvm  v% ÞT N ðvm  v% Þ; ð7Þ
v

where the subscript m represents the measured set of DOFs. Subtracting Eqs. (2) and (6) we
obtain the equation that relates the MECE objective with the general objective (5):
Df ¼ K ðv  uÞ: ð8Þ
Considering both conditions (5) and (7) in one single hybrid objective function and using Eq. (8)
we obtain
* ðvm  v% Þ;
min ðv  uÞT K ðv  uÞ þ a ðvm  v% ÞT K ð9Þ
v

with the static flexibility matrix for the equilibrium term Y ¼ K1 and the reduced stiffness matrix
for the regularisation term X ¼ K e : A variation is to express the expanded mode shape as a linear
combination of certain number of mode shapes of the model:
v ¼ Uq; ð10Þ
where U has nn modes. The system equivalent reduction expansion process (SEREP) expansion
[12] represents the experimental vector as a linear combination of certain number of numerical
mode shapes. The best estimate for the modal displacements is found by the least squares pseudo-
inverse Uþm:
qse ¼ Uþ
mv%: ð11Þ
 ARTICLE IN PRESS

128 R. Pascual et al. / Mechanical Systems and Signal Processing 19 (2005) 123–138

Replacing the reduced stiffness matrix N ¼ Kse defined by the SEREP operator in problem (9), it
may be written as
min ðu  vÞKðu  vÞ þ aðv%  vm ÞT Kse ðv%  vm Þ: ð12Þ
v

Next, problem (12) may be expressed as

min
me
ðu  vÞT Kðu  vÞ þ aðv  vse ÞT Kðv  vse Þ: ð13Þ
q

If K and M are expressed in terms of their modal decomposition [13], we obtain the following
expression for the MECE modal displacement in terms of the SEREP modal displacement:
 1 se
% 2 X2 Þ2 þ aI
qme ¼ a ðI  o q ; ð14Þ
where X is a diagonal matrix containing the numerical eigenfrequencies associated with U:

2.2. Damage assessment

The objective of the error localisation methods is to detect the location where the damage or
model errors may be present.

2.2.0. MECE indicator

From problems (5) and (8), it is straightforward to define an element level error indicator closely
related to the MECE expansion:
es ¼ ðu  vÞT Ks ðu  vÞ: ð15Þ
If several experimental modes are used in the analysis, the contribution of the mode shape v may
be normalised by
ðu  vÞT Ks ðu  vÞ
es ¼ : ð16Þ
ðu  vÞT Kðu  vÞ

2.3. Limitations of the MECE localisation

We observe several sources of problems when using the MECE indicator. They may be inherent
to the indicator associated to the experimental mode shape and due to the expansion method.

2.3.1. Excitation of model errors

The residual force can be expressed as a sum of two terms:
Df ¼ ZDv þ DZ v ; ð17Þ
where the first term on the right hand is dependent on the criterion used for the expansion and
the second one is an unknown constant. If errors are localised, the non spurious component of
the residual force is a sparse vector, almost full of zeros. Non-zero values are dependent on
the ability of the experimental mode shape v to excite the substructure(s) that present(s)
errors ðDZ Þ:
Df  ¼ DZ v c0: ð18Þ
 ARTICLE IN PRESS

R. Pascual et al. / Mechanical Systems and Signal Processing 19 (2005) 123–138 129

A potentially dangerous situation for error localisation is to identify a mode shape which does
not excite the error:
DZ v E0: ð19Þ
Modes verifying Eq. (19) are blind to model errors. It is necessary for the expanded vector to
produce a good level of deformation in the substructures where error might be present. This
problem is not addressed in this work.

2.3.2. Scattering of the residual energy

Eq. (8) establishes that the residual displacement vector v  u is the solution of the equivalent
static problem:
Kðv  uÞ ¼ Df: ð20Þ
The solution of problem (20) results in residual deformations distributed over the structure. The
damaged elements will inevitably affect the dynamic behaviour of the neighbouring elements,
because they are sharing at least one DOF. For this reason, the strain energy error indicator (15)
will show scattered residual energies.

2.3.3. Limited expansion basis

To describe the limitations of using a set of numerical mode shapes for the expansion we start
from Eq. (2):
% 2 K1 M v:
u¼o ð21Þ
If the expanded vector v is constrained by Eq. (10) and a spectral development of K1 is used, u
may also be expressed as
nfe
!
2
X fj fTj Xnn
u¼o % M qk fk ; ð22Þ
j¼1
o2j mj k¼1

where qk corresponds to the estimated modal displacement associated to fk : Eq. (22) can be
simplified to
Xnn  2
%
o
u¼ q j fj : ð23Þ
j¼1
o j

This results allows us to express the content of the residual vector v  u; whose strain energy
distribution will be used for damage assessment:
X nn  2 !
%
o
vu¼ 1 q j fj : ð24Þ
j¼1
oj

Since v  u is built up by combining a limited set of modes shapes, its energy content will be
limited to what these modes shapes are able to represent. Eq. (24) may also be written as
v  u ¼ U#q ð25Þ
 ARTICLE IN PRESS

130 R. Pascual et al. / Mechanical Systems and Signal Processing 19 (2005) 123–138

with
q# ¼ ðI  o
% 2 O2 Þq: ð26Þ
Fortunately, the richness of the experimental information provided by a laser Doppler vibrometer
(LDV) allows us to consider a great number of mode shapes in the expansion basis. Considering a
large number of mode shapes in the expansion basis, the residual forces are reproduced in a more
accurate way, therefore the damage detection will provide more accurate results.

2.4. Deciding the size of the expansion basis

Eq. (25) can be useful to select the number of mode shapes that should be in the expansion basis
U; since it needs to be wide enough to show peaks of the residual energy at the substructure where
parametric error might be present. It would be interesting to establish for each substructure s the
maximum of the residual energy it may carry:
ðv  uÞT Ks ðv  uÞ
Pm
s ¼ max ð27Þ
8 vu ðv  uÞT Kðv  uÞ
under constraint (25). The problem may be rewritten as
q# T ðUT Ks UÞ#q
Pm ¼ max ð28Þ
s
8 vu q# T ðUT KUÞ q#
which is written as the standard Rayleigh quotient [14]:
x T As x
Pms ¼ max ð29Þ
8;x xT x
with
x ¼ g1=2 q# ðiÞ ; ð30Þ

g ¼ UT KU ð31Þ
and
As ¼ g1=2 ðUT Ks UÞ g1=2 : ð32Þ
Since As is symmetric, Pm
corresponds to its maximum eigenvalue, according to the Courant–
s
Fischer Minimax theorem [14]. As ARnn nn and is easy to compute since the product ðUT Ks UÞ is
evaluated only for the degrees of freedom related to the substructure s:

2.5. Optimal sensor placement

It has been shown that the quality of the expansion process is highly dependent on the choice of
the measured degrees of freedom. If the modal MECE expansion is used, the best sensor
configuration will be the one that optimises the estimation of the modal displacements vector q of
Eq. (14) in some sense. Ref. [15] presents a technique that optimises the set-up for the basic modal
projection expansion. It is assumed that the measured vector can be decomposed in two parts:
v% ¼ Uq þ n: ð33Þ
 ARTICLE IN PRESS

R. Pascual et al. / Mechanical Systems and Signal Processing 19 (2005) 123–138 131

The degree of validity of this assumption depends on the richness of the selected modes to
generate the subspace where v% belongs to. Let us assume that noise n on the measurements
follows:
EðnnT Þ ¼ W20 ¼ j20 I;

EðnÞ ¼ 0;
where E is the expectation operator. The criterion to choice sensors will be given by the
minimisation of the covariance matrix for the estimates of the modal displacements for a SEREP
expansion:
min jj½E ðq  q Þðq  q ÞT jj;
which can be written (using Eq. (45)) as
min jjj20 ½UT U 1 jj
or equivalently
max jj½UT U jj:
This gives a criterion to select those DOFs that will be erased from the measured set: ‘‘The DOFs
which do not contribute (or very little) to the norm of jjUT Ujj are useless to be measured’’. For
convenience, the trace will be used as matrix norm
jjUT Ujj ¼ trðUT UÞ
or, exploiting the invariance of the trace:
jjUT Ujj ¼ trðKÞ;
where K is derived from the associated eigenvalue problem
ðUT UÞW ¼ KW:
Then, the EfI vector is defined by
EfI ¼ ðUWÞ2 K1 i;
where 2 indicates a term by term product, and i is a vector full of unit values. EfI outputs the
best configuration for a SEREP expansion. Given relation (14), the same can be said for the
MECE expansion (11). EfI is the optimal set-up for error localisation, when using the MECE
error localisation indicator.
According to the steps previously described, Fig. 1 describes the general procedure to define an
adequate experimental set-up that assures a given level of quality in the expansion/error
localisation.

3. Example

Let us consider the structure displayed in Fig. 2, which represents a rectangular plate with
free–free boundary conditions. A summary of the FE model properties is given in Table 1.
 ARTICLE IN PRESS

132 R. Pascual et al. / Mechanical Systems and Signal Processing 19 (2005) 123–138

Finite Element model j=j0 Modal base

mode shapes ε =ε 0 φj
j=j+1

ε0
m no
TT j

yes

Optimal sensor setup

Efi technique

Measurements
Experimental modal analysis

Expansion

Damage
end
localization

Fig. 1. Damage assessment process.

0.25
0.3

0.1 0.15
0
0 Damage here
0.1
Y axis 0.15 X axis
0.25 0.3

Fig. 2. Structure and perturbed elements.

 ARTICLE IN PRESS

R. Pascual et al. / Mechanical Systems and Signal Processing 19 (2005) 123–138 133

Table 1
Data properties for the FE model
Properties Parameters Value Unit
General Young 2:1  1011 N=m2
Poisson 0.3
Density 7800 kg=m3
Plates Elements 31  31
Thickness 0.001 m
Sides length 0:6  0:5 mm
DOFs 3072
Nodes 1024

45

40

35

30
Residual energy %

25

20

15

10

5 max
min
average
0
0 80 160 240 320 400 480 560 640 720 800
Number of modes
m
Fig. 3. P vs. the number of eigenmodes.

The purpose of the test is to simulate the experimental measurements of a damaged state of the
structure shown in Fig. 2 in order to allow us to build the expanded mode shapes. The influence
of the number of modes shapes in the expansion basis and the number of experimental DOFs
are analysed to assess a good expansion. After the decisions on the number of modes shapes
in the expansion basis and number of experimental measurements are made, the localisation of
the damaged elements may be performed. The damage is modelled as a reduction of stiffness in
the four elements highlighted in Fig. 2. The reduction is 60% of the elementary stiffness matrix.
 ARTICLE IN PRESS

134 R. Pascual et al. / Mechanical Systems and Signal Processing 19 (2005) 123–138

3.1. Deciding the number of measured DOFs and the size of the expansion basis

As it was explained in Section 2.3, the residual vector v  u will be a linear combination of nn
analytical mode shapes. Therefore, its content of strain energy distribution will give different
values for damage detection depending on what the nn numerical mode shapes are capable to
reproduce. To ensure a good level of the residual energy which an element may carry, a minimum
number of experimental DOFs must be measured and a minimum number of mode shapes must
be in the expansion basis. Fig. 3 shows the evolution of the Pm indicator as the number of
eigenmodes in the expansion basis increases. A large number of modes will produce a higher value
for the residual strain energy that an element may concentrate.
The Pm analysis is performed with an increasing number of eigenmodes, ordered according to
their natural frequencies. An alternative approach could be an optimal subset selection, leaving
only the numerical eigenmodes which allow a good representation of the experimental behaviour.
Fig. 3 shows that considering the first 400 modes in the expansion allows a good localisation, since
the Pm shows peaks up to 30% of the total residual strain energy. The convergence of the Pm
analysis is related to the form of the structure and also to the type and size of elements used in the
FE model. In this particular case, the asymptotic behaviour is reached because the experimental
structure is very simple and all the elements are equal and have the same size, except for those at
the edges of the plate. It is impossible have a 100% for the residual energy stored in an element for
the reasons explained in Section 2.3.2.
In general, it is difficult to have an asymptotic behaviour for a structure with a complicated
configuration and modelled numerically with elements of different types and sizes.

3.2. Experimental DOFs

3.3. Results

The first three expanded experimental mode shapes used in the localisation are shown in Fig. 7.
For the case of the simulated accelerometer mesh, Fig. 4(a), the expansion is done with 40 mode in
the expansion basis, since it has only 64 measurements points. The case of Fig. 4(b) is expanded

x x xx
xx x xx xx xx
x x xx x x xx xx
x x x xx x x xx xx x x x x x x x
x xx x xx x x x x x x x xx xx
xx x x x x x x xx xx x x x x xx xx x xx x x
x x x xx x x x x x x x x x
x
x
x x xx x x x xx xx xx x x x x x xx xx xx xx xx x xx x x
x xx x x x x x x x x x x
x x x x x x xx x
x x
x x xx x x xx x xx xx xx xx x x xx xx xx x x x x x x x x x x x x x x x xx x x
x x x x x x x x x x x x x x x x
xx x x x x x x x x x x x x x x x x x x x x x x x xx xx xx xx x xx x x
x x x x xx xx x x x x x x x x x xx x x x x x x x x
x x x x x x xx x xx x x x x x xx x x x x x x x xx xx x x x xx x x x xx xx xx xx xx xx xx x xx x xx x x
x x xx xx xx x x x x x x x x x x x x x x x x x x x x
x x
x
x x x x x x x xx x x x x x x x x x x x x xx xx x x x xx xx xx xx x x x x xx x x
x x xx x x x x x x xx x x x x x x x x x x x x x x x x x x x x x
0.25 xx x 0.25 x x x x x x xx x x x x x x x x x x x x x x x x x x x x x x
x x x x x x x x x xx x x x x x x x x x x x x x x x x x x x x x x
x 0.3 x x x x x x xx x x x x x x x x x x x x x x x xx x x 0.3
xx x xx x x x x x x xx x x x x x x x x x x x x x x
x xx xx x x x xx x x x x x x x
0.1 x 0.15 0.1 x x x x x x x x x x x x x xx x xx x x 0.15
x x xx xx x x xx x x x x xx x
x x x x x x x x x xx x xx x
0 x 0 x x x x x x x xx x x
xx 0 xx xx xx x 0
x xx x
0.1 0.1
0.15 0.15

Y axis 0.25 0.3 X axis Y axis 0.25 0.3 X axis

(a) (b)

Fig. 4. Sensor set-up for EfI technique: (a) 64 DOFs, (b) 600 DOFs.
 ARTICLE IN PRESS

R. Pascual et al. / Mechanical Systems and Signal Processing 19 (2005) 123–138 135

Fig. 5. MECE localisation. Simulated experimental mode #1: (a) 64 DOFs. Noise-free; (b) 600 DOFs. Noise-free;
(c) 64 DOFs. 2% noise; (d) 600 DOFs. 2% noise.

with 400 modes in the expansion basis. The reason between the number of eigenvectors in the
expansion basis and the number of experimental DOFs is 23 for the two cases. As expected a better
localisation is reached with a large number of measurements points and mode shapes in the
expansion.
The importance of the number of analytical eigenmodes in the expansion was clearly shown
with the Pm analysis. This indicator is a powerful tool to estimate a priori the localisation
limitations associated with the modal base selection. The localisation using a limited expansion
basis and less measurements point produces more scattering of the residual energy over the
structure. This situation is clearly shown with the black zones in Figs. 5 and 6. A little number of
measurements make no difference in the level of residual energy stored in an element. Thus, their
quantity is a very important matter. On the other hand a great number of DOFs allow higher
levels of elementary residual energy, see residual energy levels reached considering 600
experimental DOFs (Figs. 5(b) and 6(b)). Figs. 5(c), 5(d), 6(c) and 6(d), respectively show the
 ARTICLE IN PRESS

136 R. Pascual et al. / Mechanical Systems and Signal Processing 19 (2005) 123–138

Fig. 6. MECE localisation. Simulated experimental mode #2: (a) 64 DOFs. Noise-free; (b) 600 DOFs. Noise-free;
(c) 64 DOFs. 2% noise; (d) 600 DOFs. 2% noise.

Y axis X axis Y axis X axis

(a) (b)

Fig. 7. Simulated experimental modes (11.2 and 14:6 Hz; respectively): (a) First mode; (b) second mode.
 ARTICLE IN PRESS

R. Pascual et al. / Mechanical Systems and Signal Processing 19 (2005) 123–138 137

Table 2
Maximum residual energy on a perturbed elements
Noise-free 2% noise
Mode no. o (Hz) 64 DOFs 600 DOFs 64 DOFs 600 DOFs
1 11.2 2.0 10.3 2.0 3.8
2 14.6 2.8 15.5 2.8 14.5

localisation results when the measurements are polluted with a random noise with mean zero
and standard deviation 2% of the displacement amplitudes. This level of noise is considered
realistic for LDV measurements. Given that the model errors are located on a nodal line of mode
#1 (Fig. 7), results are quite sensitive to the existence of noise in the measurements (as described
by Eq. (19). This is not the case for mode #2 where results are much less unsensitive (see also
Table 2).

4. Conclusions

In this paper a procedure to the expansion/damage assessment process using novel concepts is
presented. The current methodology is unique among other approaches since it deals with: (1) size
of the expansion basis; (2) number of experimental DOFs; and (3) sensor placement.
A numerical example using a rectangular plate-like structure is used to show the following:
(1) The damage assessment process is quite sensitive to the number of modes shapes in the
expansion basis. In order to verify that the expansion basis is rich enough, we introduced the
Pm indicator which gives the upper bound for the residual energy which may be carried out by
a given substructure using modal projection. With this indicator the number of mode shapes
in the expansion basis is addressed to guarantee a minimum percentage of total residual
energy stored in an element.
(2) Having a great number of experimental DOFs allow us to consider a great number of
numerical eigenmodes in the expansion basis. A consequence of this fact is that the residual
force is reproduced in a more accurate way, spreading less residual energy over the structure
and detecting only the perturbed elements. Another advantage is that the expansion procedure
is reduced since the number of analytical and experimental DOFs became closer.
(3) A deficient expansion generates a smoothed expanded vector causing a residual energy
scattered all over the structure. The sources of these inconveniences are the poorly chosen
sensor set-up, the limited number of experimental DOFs and the size of the expansion basis.
The example compares two situations, overcoming the problems previously explained, by a
sensor placement method, high-spatial density measurements and an indicator for the size of
the expansion basis.
(4) The combined use of the Pm and the sensor placement method technique provides a good
expansion and therefore a suitable damage localisation.
 ARTICLE IN PRESS

138 R. Pascual et al. / Mechanical Systems and Signal Processing 19 (2005) 123–138

Acknowledgements

The authors wish to acknowledge the partial financial support of this study by the FOndo
Nacional de DEsarrollo Cient!ıfico Y Tecnologico
! (FONDECYT) of the Chilean government
(project 1020810).

References

[1] R. Pascual, Model based structural damage assessment using vibration measurements. Ph.D. Thesis, (Universit!e de
Liege), Belgium, 1999.
[2] O.S. Salawu, Detection of structural damage through changes in frequency: a review, Engineering Structures 19 (9)
(1997) 718–723.
[3] C.P. Ratcliffe, Damage detection using a modified Laplacian operator on mode shape data, Journal of Sound and
Vibration 204 (3) (1997) 505–517.
[4] K. Pandey, M. Biswas, M.M. Samman, Damage detection from changes in curvature mode shapes, Journal of
Sound and Vibration 145 (1991) 321–332.
[5] Z.Y. Shi, S.S. Law, L.M. Zhang, Structural damage localisation from modal strain energy change, Journal of
Sound and Vibration 218 (5) (1998) 825–844.
[6] U. Lee, J. Shin, A frequency-domain method of structural damage identification formulated from the dynamic
stiffness equation of motion, Journal of Sound and Vibration 257 (4) (2002) 615–634.
[7] W.X. Ren, D.J. Yu, J.Y. Shen, Structural damage identification using residual modal forces, Proceedings of XXI
International Modal Analysis Conference, Florida, 2003.
[8] A. Messina, E.J. Williams, T. Contursi, Structural damage detection by a sensitivity and statistical based method,
Journal of Sound and Vibration 216 (5) (1998) 791–808.
[9] J.T. Kim, N. Stubbs, Improved damage identification method based on modal information, Journal of Sound and
Vibration 252 (2) (2002) 223–238.
[10] W. Gawronski, J.T. Sawicki, Structural damage detection using modal norms, Journal of Sound and Vibration 229
(1) (2000) 194–198.
[11] S.K. Thyagarajan, M.J. Schulz, P.F. Pai, J. Chung, Detecting structural damage using frequency response
functions, Journal of Sound and Vibration 210 (1) (1998) 162–170.
[12] J. O’Callahan, P. Avitabile, System equivalent reduction expansion process (SEREP), Proceedings of the 7th
International Modal Analysis Conference, Nevada, 1989, pp. 29–37.
[13] M. Geradin, D. Rixen, Theorie Des Vibrations, Masson, Paris, 1996.
[14] G.H. Golub, C.F. Van Loan, Matrix Computations, John Hopkins University Press, Baltimore, MD, 1996.
[15] D.C. Kammer, L. Yao, Enhancement of on-orbit modal identification of large space structures through sensor
placement, Journal of Sound and Vibration 171 (1) (1994) 119–139.