Engineering Structures 31 (2009) 10681076

Contents lists available at ScienceDirect

Engineering Structures
journal homepage: www.elsevier.com/locate/engstruct

A direct identification procedure for assessment of stiffness distribution

Yiska Goldfeld
Faculty of Civil and Environmental Engineering, TechnionIsrael Institute of Technology, Haifa 32000, Israel

article

info

Article history:
Received 24 October 2007
Received in revised form
23 December 2008
Accepted 25 December 2008
Available online 7 February 2009
Keywords:
Identification procedure
FE-model
Curvature mode shape
Axial strain

a b s t r a c t
An identification procedure is presented for assessment of the stiffness distribution in structures with the
aid of an inverse-problem algorithm, based on an FE model of the structure with an unknown stiffness
distribution and a subset of measured vibration frequencies and vibration modes. Two independent
stiffness indicators axial and flexural are used, determined by means of the axial strain and the
curvature mode shapes, respectively. The procedure permits simultaneous location of damaged elements,
with accurate quantification of damage severity. Furthermore, it is applicable to a variety of structure
types, including frames, beams and trusses. The effects of random measurement noise and of realistic
joints are taken into consideration. The effectiveness, reliability, and range of application of the procedure
are demonstrated in a numerical study.
2009 Elsevier Ltd. All rights reserved.

1. Introduction
Damage identification procedures generally comprise three
main levels: detection, location and quantification, which provide
successively increasing amounts of information on the damage.
Therefore, if each level provides reliable information for the
next level, the first level is the most important and probably
the most difficult, since no information of the damage is
known. Most existing methods deal with the last two levels,
addressed consecutively (see for example, [1]), and a more
convenient approach would consist in simultaneous tackling of
both. Accordingly, the objective of the proposed procedure is
an algorithm that can simultaneously indicate the locations of
the damaged elements and accurately quantify their severity for
various types of structures.
Various identification techniques have been developed over
the years. The main relevant works on this topic are as follows:
Doebling et al. [2,3] and Sohn et al. [4] compiled comprehensive
literature reviews on damage identification methods. A survey of
model updating methods is found in [5,6]. Alvandi and Cremona
[7] reviewed common vibration-based damage identification
techniques, and [8] reviewed methods of damage detection
through changes in frequency.
Detection methods using changes in modal parameters fall
under two main distinct approaches: the response-based, [9],
and the model-based, [10]. The first approach compares the

Tel.: +972 4 8293044; fax: +972 4 8295697.

E-mail address: yiska@tx.technion.ac.il.

0141-0296/$ see front matter 2009 Elsevier Ltd. All rights reserved.
doi:10.1016/j.engstruct.2008.12.018

modal parameters of the undamaged structure with those of

its damaged condition, whose severity can be assessed through
the changes in the natural frequencies, mode shapes, and
damping ratio. Usually applied through optimization techniques
and sensitivity methods, it is highly effective except for one major
drawback: the parameters of the undamaged structure have to
be known, but are often unavailable in practice. In the second
approach, the objective is a set of parameters of a mathematical
model of the considered structure (in most cases an FE-model)
with a view to an optimal correlation between experimental
measurements and numerical calculations. These parameters are
then used for assessing the damage. The main drawback of this
approach is the heavy computational effort involved, but it has a
major advantage over its response-based counterpart in that only
damaged-condition data are needed (inverse-problem technique).
This advantage, combined with modern computer resources, opens
the way to new areas of application. Accordingly, the model-based
approach has been adopted in the present study.
For the proposed concept to be fully realized, several basic
obstacles still need to be overcome, namely: (a) measurement
uncertainty and inadequate test data; (b) systematic errors; (c) in
real structures damage-induced redistribution of the internal
forces and moments, with the resulting deviation from boundary
conditions and support layout of the FE model; (d) irrelevance
of the ideal joints and supports of the model in a realistic
structure. In these circumstances, many of the identification
methods reported in literature were confined to simple structures,
usually to statically-determinate beams (see for example, [11]).
This paper extends the method proposed by the author [23]
which was limited to beam structures. The proposed method is
analogous to direct calculation of the dynamic stiffness, [12]. The

Y. Goldfeld / Engineering Structures 31 (2009) 10681076

inverse-problem algorithm used in it is based on an FE-model of

the structure with unknown stiffness distribution and on a subset of measured vibration frequencies and vibration modes. Two
independent stiffness indicators flexural and axial are defined
for each element and determined by means of the curvature and
axial mode shapes, respectively. The main extension in this study
is by adding the axial stiffness indicator, therefore, to extend the
implementation of the proposed procedure to wider types of structures, such us: trusses and bar structures, frames etc. Since damage consists typically of local phenomena, their effect on the lower
frequency in the global measured response of the structure may
be insignificant. Pandey et al. [13] found that the curvature mode
shape is more sensitive to local damage. Yet, estimation of the
curvature mode shapes from experimental data is difficult and
usually involves a post-processing procedure. In this work it was
found that the axial strain mode shapes are also sensitive to local damage compared to the in-plane axial displacements, and
since the former are obtained directly from measurement data,
their use in the identification procedure makes it more promising. On the other hand, changes in the axial strains due to damage are smaller compared to those in the curvature mode shapes,
and bearing in mind the inherent measurement noise, the stiffness thus predicted is less reliable. As a solution, this paper recommends a direct procedure readily implemented in a general FE
model, practically, it can be integrated into any commercial FE
computer code and can be further used as the identification procedure in experimental tests. It involves independent recourse to
two stiffness indicators axial and flexural determined by the
corresponding mode shapes, respectively. Thus, the stiffness distribution is identified, the damage located, and its severity evaluated
simultaneously.
Another advantage of the proposed procedure is the need for a
very small number of mode shapes (in theory, even a single mode
suffices), which is the case in practical experimental tests. In order
to improve its practicability, the effects of random measurement
noises and of realistic joints are taken into consideration. A
numerical study with the examples of a plane truss and a frame
(Section 4) shows that it is able to locate both localized and
multiple damage and determine its severity with a high level
of reliability.
2. Theoretical background

K i =
i2 M i

for i = 1, . . . , n

(1)

where K and M are the analytical global stiffness and mass

matrices, respectively, usually obtained by an FE-model;
i and i
are the i-th natural frequency and i-th mode shape of the structure;
and n is the total number of obtainable mode shapes.
Since in practice the natural frequencies and mode shapes
are measured values, their substitution in the above analytical
equation will not satisfy the equilibrium condition. Besides, the
analytical boundary conditions and stiffness distribution differ
from the real ones. Therefore, the eigenvalue equation for a realistic
structure should read:
Kd i =

2
i M i

for i = 1, . . . , n

3. Solution procedure
3.1. Direct stiffness calculation
The procedure makes use of two basic relations. The first is
that the dynamic bending stiffness (EIe ) in each section is equal to
the internal bending moment (Me ) in that section divided by the
corresponding curvature (second derivative of bending mode):

(2)

where i and i are the measured counterparts. M is again

the analytical mass matrix (a common assumption in damage
identification procedures). Kd is the (unknown) counterpart of K
in Eq. (1).

Me

EIe =

(3)

The second is that the axial stiffness (EAe ) in each section is equal
to the internal axial force (Nxxe ) in that section divided by the
corresponding axial strain:
EAe =

N xxe

xxe

(4)

The dynamic torsion stiffness (GJe ) in each section is similarly

obtainable by dividing the torsional moment (T ) by the corresponding torsion rate, see [12].
3.2. Calculation of the curvature (e ) and axial strain (xxe )
Calculation of the curvature (first and second derivatives)
from measured mode-shapes, e.g. by using central-differences
approximation, is extremely sensitive and yields fluctuating and
inaccurate results. With FE as the numerical model, the second
derivative of the displacement functions is obtainable directly from
the elements shape functions. The normal displacement, along the
element, is related to the following four degrees of freedom:

ue =

i (x) u(i e)

(5)

i=2,3,5,6

where i (x) denotes the displacement functions of the element

(e)
due to a unit displacement ui , and the curvature is:

X d2 i (x)

e =

i=2,3,5,6

dx2

(e)

ui

(6)

(e)

where ui are given from the measured mode-shape. The elements

can be chosen so that nodes coincide with the measurement points.
The axial strain is defined by:

xxe =

The modal characteristics of a linear undamped system are

described by the following analytical eigenvalue equation:

1069

(7)

le

where e is the difference of the modal axial displacements

(e)
(e)
between the ends of the element (e = u4 u1 ), and le is the
elements length.
3.3. Calculation of the internal modal forces and moment
By virtue of the assumption in Section 2, the right-hand side of
Eq. (2) is known and the system can be seen as pseudo-static: for
each mode, the internal forces and moments are due to the internal
load which can be calculated as the product of local mass and local
acceleration, [12]. The k-th component of the pseudo-static load,
P(i) (for each mode i), can be written as:
(i)

Pj

2
i

ndf Z
X
k=1

Le

(k)

m (x) k (x) j (x) dx i

(8)

(k)

where i is the k-th degree of freedom of the measured mode

shape, and ndf is the total number of degrees of freedom in the
beam, and Eq. (2) becomes:
Kd i = P(i) .

(9)

1070

Y. Goldfeld / Engineering Structures 31 (2009) 10681076

Fig. 2. Degrees of freedom for smoothing process.

structure), and eiter is the vector of displacements and rotations of

the e-element given by the global equation:
1 iter
Kiter
= P.
d

(13)

Then, the dynamic stiffnesses in each element are calculated by:

EIeiter =
EAiter
=
e

Fig. 1. Iterative scheme for calculation of the stiffness distributions.

The only unknown variables in Eq. (9) are the coefficients of the
stiffness matrix.
As explained in the Introduction, the proposed procedure
resorts to two indicators. The flexural indicator, eEI , is related to
the reduction in the bending stiffness in each element, and is
defined as:

eEI =

EIed
EIeu

(10)

and the axial indicator, eEA , is related to the reduction in the axial
stiffness, namely:

Meiter

(14)

iter
Nxx
e

xxe

(15)

The curvature e and strain xxe are calculated directly from the
measured mode shapes.
On the one hand, the stiffness distributions are conveniently
identified via a single mode shape; on the other, the mode shapes
and axial strains appear in the denominators and may vanish along
the structure, so that inaccurate results would be obtained for
those elements. Therefore, the procedure first identifies, for each
given mode shape, the elements in which the curvature or axialstrain mode shapes vanish, and then uses other modes to identify
the stiffnesses in question. Practically the procedure eliminate
results in which the curvature or axial-strain mode shapes are
lower than a certain value compare to other results obtain for each
given mode.
3.4. Smoothing the measured mode shapes

EA
e

EAde
EAue

(11)

where EIeu , EIed and EAue , EAde represent the respective analytical
and predicted stiffness. The ranges of the indicators are 0 <
eEA , eEI < 1. In order to simplify validation of the obtainable
stiffness estimates, and since the proposed procedure is applied
on a linear system, the indicators are calculated independently
in each iteration. Such treatment of the damage guarantees that
the resulting stiffness matrix is positive definite, and imposes
connectivity of the elements in the structural mode.
In order to reduce the number of unknown parameters, a
representative pattern of the reduction in the stiffness matrix due
to cracks can be used, see [1421].
For the case of a statically indeterminate structure the following
iterative process is applied to the numerical model (see Fig. 1):
The modal internal forces and moments (Niter
e ) are determined
in each element by solving the local element equation:
1 iter
Niter
= d kiter
e
e
e

(12)

where Niter
e consists, for the case of truss element: the internal axial
forces, and for the case of frame element: the internal axial and
1
shear forces and bending moment. d kiter
is the damaged stiffness
e
matrix of the e-element, which is known from the previous
iteration depend on the chosen element type (the first iteration
can use the stiffness distributions of the analytically undamaged

The measured mode shapes are obtained in vibration tests.

As these always involve some level of noise and inaccuracies,
smoothing is called for. Optimal smoothing is expected to reduce
the noise effect without detriment to the measurement accuracy
of local damage. There are two main approaches to smoothing
global and local. Under the global approaches the measured
deflection is fitted to a polynomial function by a least-squares
technique. Its disadvantage is a risk of deletion or distortion of the
damage effect, which is mostly local. The local approach, proposed
by Maeck and De Roeck [12] involved a so-called weighted
residual penalty-based technique, and is adopted here with a few
improvements. Its main principle lies in dividing the structure
into a number of segments separated by nodes corresponding to
the measurement points. Each node has two degrees of freedom
corresponding to the modal normal displacement v and the modal
rotation , which are approximated independently. Linear shape
functions Ni are used, see Fig. 2. The objective function ( ), which
has to be minimized, contains the difference of the approximated
and measured mode shapes with an additional penalty term
imposing continuity of rotation in an average smoothed way:

v md

L2e
2

2

dx +

L2e
2

2
Z 
dv

dx
dx

mr

2

dx

(16)

Y. Goldfeld / Engineering Structures 31 (2009) 10681076

1071

Fig. 3. Geometry of plane truss and damage scenarios: (a) single damaged member,
(b) two damaged members, (c) three damaged members.
d
r
where m
and m
denote the measured normal displacement and
rotation mode shape, respectively, and Le is the length of the
segment. The two first terms indicate that the average difference
has to be minimized. (If rotations are not available from measured
mode shapes, = 0, otherwise = 1). The last term filters the
measured errors and thus smoothes the deflection and rotation
curves. The weight of this extra condition is set by the chosen
dimensionless penalty factor .
Deriving the objective function Eq. (16) for the unknown
degrees of freedom, substituting the shape function and solving
the integral expressions, an analytical form is obtained for the
governing system on segment level:

Le
3

+ Le

Le 3
3

Le 2
2

Le
6

Le 3
3

Le
3

symm.

Le
Le 2
2

+ Le

md 1 Le
md 2 Le

6
3

r
r
e
3
e
3

m 2 L
m 1 L

+
6

+

6
6 1
e2
v2
L

2
2

Le 3
Le 3

+
Le 3

d Le
md 2 Le
m1

3
6

r
r
e3
e3

m2

m 1

Le 2

Le 3
v1

(17)

Assembling the above matrix in the global, we obtain a symmetric,

non-singular system of the same order as the original stiffness
matrix. After smoothing of the measured mode-shape, the
identification procedure can be applied.
It should be noted that here the measurement noises in the axial
mode shapes are not smoothed, so that the latter can be used as
measured in the identification procedure.

Fig. 4. Axial indicators obtained with distorted modes (error = 5%), for truss with
single damaged member (no. 6).

4. Numerical examples
In order to study the effect of measurement noise on the
accuracy of the proposed procedure, the analytical mode shapes
were distorted with random noise. The distorted signal is
represented, see [22], as:





ij = ij 1 + i  max,j 

(18)

where ij and ij are the mode shape components of the i-th

mode at the j-th degree of freedom with noise and without noise,

respectively; i is a random number with zero mean and variance

1 (normal distribution); is the random noise level; max,i is the
largest component in the i-th mode shape.
Since in this study the noise is evaluated randomly (Eq. (18)),
the presented numerical study concerns a specific case of random
distribution. However, the author has verified the proposed
algorithm on various random cases.
4.1. Plane truss
The steel truss of the example is shown in Fig. 3. The geometrical
and material properties were taken from [22], as follows: E =
200 GPa, = 7.8 103 kg/m3 , L = 1 m and A = 0.004 m2 .
4.1.1. Perfect hinge joints
For the case of perfect hinge joints, three damage patterns are
examined involving a sequence of numbers of damaged members:
one (no. 6), two (nos. 5 and 6), and three (nos. 4, 5, 6) which count as

1072

Y. Goldfeld / Engineering Structures 31 (2009) 10681076

Fig. 5. Axial indicators obtained with distorted modes (error = 15%), for truss with
single damaged member (no. 6).

Fig. 6. Axial indicators obtained with distorted modes (error = 15%), for truss with
two damaged members (nos. 5, 6).

multiple. All damaged members underwent a stiffness reduction

of 20%.
The first damage pattern is investigated with mode shapes
distorted by 5% and 15% random noise. Results of the axial stiffness
indicator are given in Figs. 4 and 5, respectively. The first iteration
assumes an analytical undamaged stiffness distribution. The axial
strain mode shapes vanish as follows: member no. 7, first mode;
no. 10, second mode; nos. 6, 8 and 13, third mode; and nos. 7
and 15, fourth mode. Accordingly, in each mode the procedure
overlooks the axial stiffness of the above members.
The norms of the relative errors (EA ) of the procedure for each
mode and for the averaged results are also given in the figures, they
are calculated by:

by 15% random noise. The reliability of the procedure is again

demonstrated, the differences between the predicted averaged
axial results and the real ones being around 2%.
For all three patterns convergence was quite fast (usually
after 5 iterations), more so in the second and third than in the
first. Application of analytical mode shapes (with no noise) yields
exactly the correct stiffness distribution in all three cases. The
author has verified the procedure for wide range of stiffness
reduction; for analytical mode shape the procedure predicts the
accurate stiffness distribution even for minor stiffness reduction
(less than 5%), for distorted mode shape the reliability of the
procedure depends on the ratio between the accurate stiffness
reduction and the random noise.

EA

v
uN 
2
el
X
EAn EAdn
1 u
t
=
d
Nel

n=1

EAn

(19)

where EAn , EAdn are the predicted and the accurate axial stiffnesses,
respectively, at the n-th element. It is seen that the procedure
yields reliable predictions of the axial stiffness indicator and
thereby of its distribution, the differences between the predicted
averaged results and the real ones being 0.5%2.5%. The reliability
is mainly influenced by the relative locations of the damage and of
the vanished axial strain. For members with the latter effect, results
in the mode in question are expected to be less reliable, as can be
seen in the third mode where the axial strain vanishes in member
no. 6.
The predicted axial stiffness indicators for the second and third
damage patterns are given in Figs. 6 and 7, respectively. Both
damage patterns were investigated with mode shapes distorted

4.1.2. Realistic joints

In studying the effect of realistic joints on the reliability
of the proposed procedure, all joints are assumed to be fully
connected, i.e. with three degrees of freedom for each joint (two
displacements and rotation).
Convergence results in the case of analytical mode shapes
for the three patterns are given in Fig. 8. It is seen that with
realistic joints convergence is quite slow; for the second and
third damage patterns more than 200 iterations were needed.
Again, application of analytical mode shapes yields exactly the real
stiffness distribution.
For the case of mode shapes distorted by 10% random noise,
the predicted axial stiffness indicators in each member are given
in Figs. 911, respectively. The axial strain mode shapes vanish at
the following members: nos. 9, 11 and 13, first mode; no. 7, second
mode; nos. 9, 11 and 12, third mode; nos. 3 and 11, fourth mode.

Y. Goldfeld / Engineering Structures 31 (2009) 10681076

Fig. 7. Axial indicators obtained with distorted modes (error = 15%), for truss with
multiple damage (nos. 4, 5, 6).

1073

Fig. 9. Axial indicators obtained with distorted modes (error = 10%), for truss with
realistic joints and single damaged member (nos. 6).

predicted damage is heavier than the real one, so that the model
results can be considered on the safe side.
4.2. Plane frame

Fig. 8. Convergence of axial stiffness for truss with realistic joints for different
damage patterns.

Again, as in the case of perfect joints in each mode the procedure

overlooks the axial stiffness of the above members.
The norms of the relative error (EA ) of the procedure for each
mode and for the averaged results are also given in the figures.
The differences between the predicted averaged results and the
real ones are 3%12%, and even though the relative errors are
quite high, in all patterns the damaged members were located with
a high level of reliability. Moreover, in the presented cases the

The second example is a statically indeterminate reinforced

concrete plane frame (Fig. 12). The material and properties are as
follows: concrete type C-30, = 2.4 ton/m3 , A = 0.08 m2 , I =
0.00106 m4 . Again, the first 4 modes are used.
Three damage patterns are examined. The first is a single
damaged segment no. 7, the second continuous damage in nos.
911, and the third multiple damage in nos. 67 and 12. In this
study, damage in the structure is equivalent to reduction of the
elements height. Therefore, reduction in bending stiffness will be
higher (order three) than reduction in axial stiffness. All damaged
segments underwent a bending stiffness reduction of 30% and an
axial stiffness reduction of 11.2% (equivalent to a 11.2% reduction
in height).
The averaged results of the predicted stiffness indicators for
the first damage pattern are given in Fig. 13 with mode shapes
distorted by 10% random noise. The distorted mode shapes are
smoothed with a weighted factor of = 0.001. The norms of the
relative error of the averaged results (EA ) and (EI ) are included
in the figure, EA is given in Eq. (19) and EI is calculated by:

v
u

EI =

2
Nel 
X
1 u
EIn EInd
t

Nel

n=1

EInd

(20)

where EIn , EInd are the predicted and the accurate bending
stiffnesses, respectively, at the n-th element. It is seen that in

1074

Y. Goldfeld / Engineering Structures 31 (2009) 10681076

Fig. 10. Axial indicators obtained with distorted modes (error = 10%), for truss
with realistic joints and two damaged members (nos. 5, 6).

terms of the flexural indicator, the procedure succeeded both in

locating the damaged segment and in evaluating the severity of
the damage with an equally high level of reliability, while in terms
of its axial counterpart the location reliability was higher than
that of damage evaluation. Noticed that even though reductions
in axial stiffness are in the same order of magnitude of the
random noise, good prediction obtained from the axial indicator.
As was mentioned before, since the algorithm is applied on a
linear system, it evaluates each indicator independently, so that
random measurement noises in one indicator play almost no
role in the other. Furthermore, since in this study damage is
equivalent to reduction in the cross section height, to be consistent,
both reduction in axial and bending stiffnesses must be taken in
consideration in frame structure even though axial indicator gives
less reliable results.
In addition to the above, a separate parametric study for the
case of unknown rotation mode shapes that is, where only
deflections are measured ( = 0 in Eq. (16)) is performed
for the first damage pattern. In Fig. 14 the predicted indicators
are given from analytical and from distorted (10% random noise)
mode shapes. In order to obtain the rotations from the normal
displacement for both mode shapes, smoothing was applied with
= 0.00001. The norms of the relative errors are included in
the figure. Here again the procedure succeeded in identifying the
flexural indicator with a level of reliability superior to that of its
axial counterpart. The location of the axial stiffness reduction was
predicted with a higher level of reliability than its severity. For both
cases the predicted flexural indicators are lower, and their axial
counterparts higher, than the real ones.

Fig. 11. Axial indicators obtained with distorted modes (error = 10%), for truss
with realistic joints and multiple damage (nos. 4, 5, 6).

The second and third damage patterns are given in Figs. 15

and 16, respectively. The mode shapes were distorted by 10%
random noise. The procedure yields reliable predictions of the
axial indicator. The differences between the predicted averaged
results and the real ones are 2.5%4.5% for the flexural indicator
and around 9% for its axial counterpart. Even though the relative
errors are again quite high, in both damage patterns the damaged
segments were located with a high level of reliability. Also the
predicted damage is again heavier than the real one, so that the
model results can be considered on the safe side.
5. Conclusion
A direct identification procedure for the stiffness distribution
of a general structure is presented. The algorithm makes use of
an FE model and a subset of measured vibration frequencies and
vibration modes. The procedure permits simultaneous location
of damaged elements, with accurate quantification of damage
severity. It is verified for distorted mode shapes and for realistic
joints. It uses two independent stiffness indicators flexural
and axial. Prediction of the bending stiffness distribution is more
reliable than that of its axial counterpart. The damage severity
predicted by the axial stiffness is mostly higher than the real one,
so that the model results can be considered on the safe side.
In theory, a single measured mode shape suffices for the
procedure, but in view of the strong dependence on the axial strain
and on the curvature mode shapes, inaccuracies are expected at
elements in which they vanish. Therefore, several mode shapes are
needed in order to identify the stiffness distribution with a high
level of reliability.

Y. Goldfeld / Engineering Structures 31 (2009) 10681076

1075

Fig. 14. Indicators obtained with analytical and distorted modes (error = 10%) and
unknown rotations, for plane frame with single damaged member (no. 7).

Fig. 12. Geometry of plane frame and damaged scenarios: (a) single damage, (b)
continues damage (c) multiple damage.

Fig. 13. Indicators obtained with distorted modes (error = 10%), for plane frame
with single damaged member (no. 7).

Fig. 15. Indicators obtained with distorted modes (error = 10%), for plane frame
with two damaged members (nos. 9, 10, 11).

Fig. 16. Indicators obtained with distorted modes (error = 10%), for plane frame
with multiple damage (nos. 6, 7, 12).

References
Acknowledgments
This work was supported by a financial contribution of Bank
Hapoalim in Israel. The support is gratefully acknowledged. The
author is indebted to E. Goldberg for editorial assistance.

[1] Wang X, Hu N, Fukunaga H, Yao ZH. Structural damage identification using

static test data and changes in frequencies. Eng Struct 2001;23:61021.
[2] Doebling SW, Farrar CR, Prima MB, Shevitz DW. Damage identification and
health monitoring of structural and mechanical systems from changes in their
vibration characteristics: A literature review. Los Alamos National Laboratory
Report LA-13070-MS. 1996.

1076

Y. Goldfeld / Engineering Structures 31 (2009) 10681076

[3] Doebling SW, Farrar CR, Prima MB. A Summary review of vibration-based
damage identification methods. Shock Vib Digest 1998;30(2):91105.
[4] Sohn H, Farrar CR, Hemez FM, Shunk DD. A review of structural health
monitoring literature: 19962001. Los Alamos National Laboratory Report LA13976-MS. 2003.
[5] Mottershead EJ, Friswell MI. Model updating in structural dynamics: A survey.
J Sound Vib 1993;167(2):34775.
[6] Friswell MI, Mottershead EJ. Finite element model updating in structural
dynamics. Dordrecht: Kluwer Academic; 1995.
[7] Alvandi A, Cremona C. Assessment of vibration-based damage identification
techniques. J Sound Vib 2006;292(1):179202.
[8] Salawu OS. Detection of structural damage through changes in frequency: A
review. Eng Struct 1997;19(9):71823.
[9] Natke HG. Error localization within spatially finite-dimensional mathematical
models. Comput Mech 1991;8:15360.
[10] Vepa K. Optimal identification of vibrating structures. In: Proceeding of the
international modal analysis conference. 1984. p. 96105.
[11] Kim JT, Ryu YS, Cho HM, Stubbs N. Damage identification in beam-type
structures: Frequency-based method vs. mode-shape-based method. Eng
Struct 2003;25:5767.
[12] Maeck J, De Roeck G. Dynamic bending and torsion stiffness derivation from
modal curvatures and torsion rates. J Sound Vib 1999;251(1):15370.
[13] Pandey AK, Biswas M, Samman MM. Damage detection from changes in
curvature mode shapes. J Sound Vib 1991;145(2):32132.

[14] Christides S, Barr ADS. One dimensional theory of cracked BernoulliEuler

beams. Int J Mech Sci 1984;26:63948.
[15] Abdel Wahab MM, De Roeck G, Peeters B. Parameterization of damage in
reinforced concrete structure using model updating. J Sound Vib 1999;228(4):
71730.
[16] Cerri MN, Vestroni F. Detection of damage in beams subjected to diffused
cracking. J Sound Vib 2000;234(2):25976.
[17] Maeck J, Abdel Wahab M, Peeters B, De Roeck G, De Visscher J, De Wilde WP,
et al. Damage identification in reinforced concrete structures by dynamic
stiffness determination. Eng Struct 2000;22:133949.
[18] Friswell MI, Penny JET. Crack modeling for structural health monitoring. Struct
Health Monitor 2002;1(2):13948.
[19] Sinha JK, Friswell MI, Edwards S. Simplified models for the location of cracks
in beam structures using measured vibration data. J Sound Vib 2002;251(1):
1338.
[20] Ren WX, De Roeck G. Structural damage identification using model data. I:
Simulation verification. J Struct Eng 2002;128(1):8795.
[21] Ren WX, De Roeck G. Structural damage identification using model data. I: Test
verification. J Struct Eng 2002;128(1):96104.
[22] Liu JK, Yang QW. A new structural damage identification method. J Sound Vib
2006;297(4):694703.
[23] Goldfeld Y. Identification of the stiffness distribution in statically indeterminate beams. J Sound Vib 2007;304(35):91831.