Journal of Sound and Vibration 340 (2015) 126137

Contents lists available at ScienceDirect

Journal of Sound and Vibration

journal homepage: www.elsevier.com/locate/jsvi

The use of modal curvatures for damage localization

in beam-type structures
J. Ciambella a,b,n, F. Vestroni b
a
b

Advanced Composites Centre for Innovation and Science, University of Bristol, University Walk, BS8 1TR Bristol, United Kingdom
Dipartimento di Ingegneria Strutturale e Geotecnica, Sapienza Universit di Roma, Via Eudossiana 18, 00184 Rome, Italy

a r t i c l e in f o

abstract

Article history:
Received 9 May 2014
Received in revised form
10 November 2014
Accepted 28 November 2014
Handling Editor: M.P. Cartmell
Available online 22 December 2014

The localization of stiffness variation in damaged beams through modal curvatures, i.e.,
second derivative of mode shapes, is studied by exploiting a perturbative solution of the
EulerBernoulli equation. It is shown that for low order modes the difference between
undamaged and damaged modal curvatures has only one distinct peak if the damage is
localized in a narrow region. This phenomenon is independent of the presence of
experimental noise and of the technique used to reconstruct the curvature mode shapes
from the displacement mode shapes. Broader damages cause the modal curvature
difference to have several peaks outside the damage region that could result in a false
damage localization. The same effect is present at higher modes for both narrow and
broad damages. As a result, modal curvatures can be effectively used to localize structural
damages only once they have been properly filtered. Here the perturbative solution is
used to introduce an effective damage measure able to localize correctly narrow and broad
damages and also single and multiple damages cases.
& 2014 Elsevier Ltd. All rights reserved.

1. Introduction
In damage detection techniques different quantities can be selected to be observed. The majority of approaches use the
variations of natural frequencies: they are easily measurable with high accuracy, but as global quantities are less sensitive to
concentrated damages [15]. In comparison, the advantage of using mode shapes and their derivatives as a basic feature for
damage detection is a well-established result. First, mode shapes contain local information, which makes them more
sensitive to local damages and enables them to be used to detect multiple damages or when nonlinearities are present in the
structural response [69]. Second, the mode shapes are less sensitive to environmental effects, such as temperature, than
natural frequencies [1012].
Over the years several damage identification techniques have exploited the use of changes in Modal Displacements
(MDs) between the undamaged and the faulty system. Such techniques have been applied with different degrees of success
to structures such as rods [13], beams [14,15], plates and even more complex systems (see, for instance, [1618]). Their
major limitation is that, even if the damage is localized, its effect on MDs can spread all over the structure. As a consequence,
it can be cumbersome to identify the actual damage position when several peaks in the MD difference appear.

Corresponding author at: Dipartimento di Ingegneria Strutturale e Geotecnica, Sapienza Universit di Roma, Via Eudossiana 18, 00184 Rome, Italy.
E-mail addresses: jacopo.ciambella@uniroma1.it (J. Ciambella), vestroni@uniroma1.it (F. Vestroni).

http://dx.doi.org/10.1016/j.jsv.2014.11.037
0022-460X/& 2014 Elsevier Ltd. All rights reserved.

J. Ciambella, F. Vestroni / Journal of Sound and Vibration 340 (2015) 126137

127

On the contrary, Modal Curvatures (MCs) are widely considered a superior indicator of damage position, as for localized
damages, MC difference is thought to be concentrated in the damaged area [1925]. In [10] it is shown that the MC shapebased methods are much more robust and promising than the MD shape-based methods, mainly when curvatures are
directly measured from the strain mode shape. However, some authors [20,26] pointed out that MCs fail to provide a clear
localization of the damage position even in simple cases such as cantilevered beams. Montalvao [26] for instance state that
higher derivatives are more promising for damage identification, but [] false damage indications may be observed at
mode shape nodal points or where the quality of the measurements is relatively poor.
One of the first attempts to identify damages through MC changes was presented in [19]. Through numerical examples
on a simply supported beam, they show that MCs are mainly localized in the region of damage compared to the changes in
the MDs. However, major oscillations appeared when they compared changes in MCs for higher modes. They also stated that
a relatively dense measurement grid is required in order to get a good estimate of the curvatures for higher order modes.
Differences in MCs have been applied in [20] to detect damage in bridges. The authors comment on the numerical
simulations in [19]: the difference in MC between the intact and the damaged beam showed not only a high peak at the
fault position but also some small peaks at different undamaged locations for the higher modes. This can cause confusion to
the analyst in a practical application in which one does not known in advance the location of faults. They surmised that the
incorrect damage position was due to the numerical scheme used in Pandey et al. [19] to calculate MCs from MDs. In trying
to overcome this problem, the authors computed modal curvatures by two different algorithms: from the MDs, by using a
central difference approximation, and from the bending moment obtained through finite element. As a result, they found
that in any case the difference in MCs shows several peaks not only at the position of the damaged element but also at other
positions regardless which algorithm was used to evaluate the MCs. In this respect, more advanced filtering techniques such
as spline interpolation [18], modified Laplace operator [27] and wavelet transform can be applied to obtain a robust estimate
of the MCs from the noise corrupted MDs; however, as shown in the next sections, the presence of peaks in the MC
difference outside the damage region is independent of the noise presence yet could affect noise free data.
Lestari et al. [24] computed the eigenvectors and eigenvalues of a damaged beam through a perturbation method. Their
solution was later corrected and improved [28]. In all these papers first-order eigenmodes contain a term depending on the
undamaged mode shape; indeed, it is shown later in Section 2 that this term has to be filtered out when considering
differences between normalized MCs or MDs. As a consequence, their filtering procedure does not provide a clear damage
localization as the resulting MC variations have relevant peaks not only in the damaged area but all over the structure. This
result is apparently in contrast with the numerical simulations carried out by several authors [19,20,23,25] where, at least
for low order modes, the main peak on the MC difference occurs in the damaged region.
A special mention deserves non-baseline techniques based on MCs which have shown promising results [see for instance
[29]]; however such techniques normally require large number of measurement points which makes their application to
large structures difficult.
In this paper, by using a perturbative solution, it is demonstrated that MCs do not convey localized information on the
damage position, if not properly processed. As a consequence, a novel filtering procedure for MCs is introduced that in turn
leads to an effective damage localization. By considering numerical examples we show that, for high order modes, the MC
difference is unsuitable for damage localization. We remark that this issue is not due to the algorithm used to computed the
MCs, as surmised by some authors, but it is intrinsic of the normalized MC difference. The proposed filtering procedure is
used to obtain from the MC difference an accurate estimate of single and multiple damage shapes by using only the
minimum number of information conveyed in the first few eigenmodes.
2. Modal analysis of damaged beam
In this section we derive a perturbative solution of the EulerBernoulli beam equation with non-constant stiffness; such
a solution is then used to obtain a closed form expression of the MC difference and introduce a novel filtering procedure for
damage localization.
Within the framework of the EulerBernoulli beam theory, we assume the damage to be a localized stiffness variation
expressed as
EIx EI 0 1  x

(1)

where E is Young's modulus of the material and I0 is the second moment of area of the undamaged beam; x is the damage
shape, i.e., a smooth continuously differentiable function that is non-zero only in the damaged region, 0 r jxj r1, and is
the intensity of damage, 0 r jjo 1.
The governing equation for the ith transverse mode of displacement vn x of the damaged beam is
"
#
4
2
2
d vni x
d vni x
d
n


(2)
 i vni x 0
4
2
2
dx
dx
dx
where vni and i are the ith eigenfunction and eigenvalue of the damaged system, respectively. For small damage intensities,
they can be expanded as a power series of , i.e.
n

vni x v0i x  v1i x o2 ;

ni 0i  1i o2

(3)

128

J. Ciambella, F. Vestroni / Journal of Sound and Vibration 340 (2015) 126137

where the dependence of i on the geometrical and constitutive properties of the beam is well known, i.e., i 2i =EI 0 ,
with i being the ith eigenfrequency of the undamaged system and is the mass per unit length.
By taking into account the contributions at different orders, the following ODE system is obtained:
0

d v0i x

0th order:

dx

1st order:

d v1i x
4

dx

 i v0i x 0
0

 i v1i x i v0i x 
0

(4)
d i x
2

(5)

dx

where the function i x is the damage shape weighted with the ith MC, i.e. i xx d v0i =dx , and Eq. (4) is the governing
equation of the undamaged system.
The complete solution of the first order equation (5) in terms of the orthogonal modes v0k x is
2

v1i x v0i x ik v0k x

(6)

kai

and from Eq. (3), the first-order approximation of the ith damaged mode shape becomes
vni x 1  v0i x  i
kai

The coefficients

v0k x o2 :

(7)

ik and the first-order eiqenvalue 1i are obtained by substituting (6) into (5) (see Appendix)
1i
ij

1
; w0
v0i 2 i i
1

v0j 2 i  j
0

(8)

i ; w0j

(9)

For Eqs. (8)(9) to hold, the damage function and its derivatives must be zero at both beam ends, i.e., 0
L 0 0 0 L 0. For the sake of brevity, the damaged and undamaged MCs have been represented
as wni x and
RL
w0i x respectively, while angle brackets have been used to indicate the scalar product, i.e., u; v 1=L 0 uv dx, while  is
used for the norm, i.e., v2 v; v.
3. Relationship between MC difference and damage shape
Standard damage identification techniques are based on the difference between normalized undamaged and damaged
MDs or MCs. From Eq. (3), the normalized damaged mode shape v ni vni =vni can be expressed as
!
 
v1i x
n
0
0
 v i x o 2
v i x v i x 
v0i
v 0i x  ik
kai

 
v0k 0
v x o 2
v0i k

(10)

which yields the following MDs difference:

i xv 0i x  v ni x ik
kai

v0k 0
v x
v0i k

(11)

where a superimposed bar has been used to indicate the normalization.

Eq. (10) states that the ith normalized MD is composed of the undamaged mode v 0i together with the contributions of the
other modes weighted through the damage dependent coefficient ij. We ought to remark that (10) is different from the
first-order solution used in [28] as in their case the undamaged MD v 0i is not removed from the first-order normalized mode
shape v1i x=v0i . As a consequence, when they compare normalized MDs or MCs obtained experimentally, peaks outside
the damage region could appear.
By using (6) and (9) together with (11), the MD difference is obtained

i x

0
0
kai i  k

i ; w 0k v 0k x

(12)

i ; w 0k w 0k x:

(13)

or equivalently the MC difference

i x

0
0
kai i  k

J. Ciambella, F. Vestroni / Journal of Sound and Vibration 340 (2015) 126137

129

Indeed, w0k x being a complete base, the damage shape i can be computed by a standard projection over the MCs as
1

k1

0k

i x

i ; w 0k w 0k x

(14)

where the equivalence k w0k is used (see Appendix). The scalar product in (14) can be evaluated by multiplying Eq. (13)
0
0

by w 0k x for i a k, i.e, i ; w 0k i =k 1i ; w 0k .

It is therefore possible to introduce the following filtering equation:
0

i x  i x

0i  ni 0
0

w i x 0i 2 i ; w 0k w 0k x
0
i
k a ik

(15)

which expresses the ith weighted damage shape i x i x in terms of MCs and their difference. Once the
0
n

eigenfrequencies, i , i , and the MC difference i are obtained from experiments, Eq. (15) provides a way to filter the
0
02
contributions of higher modes and recover the ith damage shape i x. Because of the term i =k , the effects of higher
order terms can be neglected in the infinite sum, which can be truncated at i 1. Therefore, low order MC differences can be
used more effectively for damage localization as they require a lower number of modes to be filtered.
It is also useful to examine the analytical expression of the ith MC difference that is obtained from Eq. (13)

i x  i x

1i 0
0
w i x i0 ik w 0k x
0
i
k a ik

(16)

which is the sum of three contributions: the ith damage shape, the ith MC and a term taking into account the contribution of the
other MCs. As Eq. (15) shows, the contribution of high order MCs can easily overcome the damage shape function giving erroneous
indication of damage also in very simple cases. This result is confirmed by the numerical simulations carried out in Pandey et al.
[19] and Abdel Wahab and De Roeck [20], where it is shown that peaks outside the damage region increase for a higher mode
number. However, Eq. (16) shows that this phenomenon is not due to the numerical error introduced with the computation of MCs

from MDs, as surmised by some authors, but it is due to the presence of the other terms in the expression of i .
No assumptions have hitherto been made upon the particular form of the damage shape x, thus the application of
Eq. (15) allows the reconstruction of i x for either single or multiple damages. This reconstructed quantity is the damage
shape weighted through the ith MC. As such, for multiple damages, the ratio between the damage intensities could be
different from the actual value according to the intensity of ith MC in the damaged region.
4. Examples
The advantages of the proposed filtering procedure are assessed against the existing localization techniques for two
prototype problems: a simply supported beam already dealt with in [19,20], and a cantilever beam as in [24]. The damage
shape identified through Eq. (15) is compared to the localization provided by the normalized MCs difference as proposed by
[19]. Other localization techniques, i.e., COMAC (Coordinate Modal Assurance Criterion) [1] and the Damage index defined in
[14], have not been able to provide a clear localization even in the simple examples considered thus they are not
reported here.
The damage shape is parameterized with a cubic polynomial to guarantee the continuity of the second derivative across
the beam length, i.e.
8
1
>
2
>
> 3 b 2x 2x0 b 4x 4x0 ; 0 o x0  b=2 rx o x0
<
b
(17)
x 1
>
2
>
>
: 3 b 4x 4x0 b  2x 2x0 ; x0 rx r x0 b=2 o L
b
as sketched in Fig. 1. In (17) x0 is the center of the damaged area and b its width. This damage parametrization is used in the
next section to derive a closed form expression for the MCs difference through Eq. (16). However, Eq. (15) is independent of

0 x0

b
2

x0

x0

b
2

Fig. 1. Damage shape function x with center x0 and width b (see Eq. (17)).

130

J. Ciambella, F. Vestroni / Journal of Sound and Vibration 340 (2015) 126137

the particular damage shape provided that x is differentiable twice and the function and its derivatives vanish at the
boundary. Therefore, the filtering procedure can be applied unchanged when x has a different form including the case of
multiple damages.
4.1. Simply supported beam
The zeroth and first-order equations (4) and (5) are solved in a closed form for simply supported beams and the
analytical expression of the MCs difference can be used to assess the sensitivity to damage position and width. The simply
supported boundary conditions for the zeroth-order problem are
v0i 0 0;

v0i L 0;

w0i 0 0;

and eigenvalues and MCs are obtained accordingly



i

L
0
i

4
w 0i x

s
2

w0i L 0

(18)

i2 2 sin i x

L5

(19)

In this case, the MCs difference (16) is

s

 x
 x
 x

L5 i
~
i x; x0 ; b
x sin i  i x0 ; b sin i  ik x0 ; b sin k
L
L
L
2 2 i2
kai

(20)

where two coefficients i x0 ; bi x0 ; b=i and i k x0 ; bi=k2 ik x0 ; b have been introduced; for the cubic damage
1
~ is the MC
shape (17), the values of i, ik, ij and q
the
first-order eigenvalues i are given in the Appendix. In Eq. (20),

i
difference normalized with respect to 2=L5 i2 2 to have the maximum of the damage shape term, i.e., x sin i =L,
equal to 1.
The first term in Eq. (20) depends on the damage shape x, thus, is non-zero in the damage region while zero
elsewhere; all the other terms can be non-zero over all the beam length. The relative weight of the coefficients i and ik
determines the intensity of peaks outside the damage region which could even exceed those in correspondence with the
damage; in this situation, the use of unfiltered MC would lead to an incorrect damage localization. Fig. 2 shows contour plots
of the functions i x0 ; b and ik x0 ; b for the first three modes (i1, 2, 3) and k o i; for k 4 i, the coefficients ik are small
compared to the other terms and can be neglected. Due to the symmetric definition of damage, the inequalities x0  b=2 4 0
1

b L

0.4
0.46

0.4

0.2
0.13
0.26

0.2
0.0
0.0

0.4 0.6
x0 L

0.17
0.3

0.4

0.26

0.8

0.0
0.0

1.0

0.2

0.4 0.6
x0 L

0.8
b L

b L

0.31

0.13
0.04

0.2

0.18

0.2

0.4 0.6
x0 L

0.27

0.18

0.09

0.8

0.0
0.0

1.0

0.15

0.12
0.03

0.2

0.4 0.6
x0 L

0.8

1.0

0.21
0.15 0.06

0.4

0.12
0.09

0.24

0.0
0.0

0.03

0.4 0.6
x0 L

0.6

0.17
0.11
0.02

0.8

1.0

0.26
0.17

0.26
0.02
0.14
0.29

0.4
0.2

0.06

0.08

0.2

0.23

0.03

0.2

0.14
0.08

0.09
0.15
0.09
0.12

0.15

1.0

0.05

0.8

0.06

0.03

0.8

32

1.0
0.18

0.12 0.31

0.6

0.2

0.09

0.0
0.0

0.21
0.06

0.15

0.27

0.22

0.4

0.31

0.27

0.24

0.2

31

0.36
0.27

0.4

0.24
0.09

0.18

0.12
0.08

1.0

0.4

0.6

0.6

0.21

0.13

0.44

0.8

0.04

0.21
0.31

0.03

1.0

0.6

0.2

0.06

0.2

0.06

0.8
b L

0.53
0.33

0.39
0.35

0.8

0.6

0.6

21

1.0

0.43

0.67

0.8

1.0

b L

1.0

b L

0.0
0.0

0.05

0.05

0.050.230.29
0.2

0.02

0.11
0.02

0.2

0.4 0.6
x0 L

0.8

1.0

Fig. 2. Contour plots with isolines of the functions i and ik for the first three modes of the simply supported beam (i1, 2, 3 and k o i) against normalized
damage position x0 =L and width b=L.

J. Ciambella, F. Vestroni / Journal of Sound and Vibration 340 (2015) 126137

131

Fig. 3. Absolute value of MCs differences (thick line) against the actual damage shapes x (thin filled line) for the simply supported beam with three
different damage cases: x0 =L 0:5 and b=L 1 for mode 1, x0 =L 0:7, b=L 0:6 for mode 2 and x0 =L 0:5 and b=L 0:57 for mode 3. In all cases, multiple
peaks outside the damage region appear.

1
0.8

Mode 1

Mode 2

0.8

0.8

0.6

0.6

0.6

0.4

0.4

0.4

0.2

0.2

0.2

0
0

Pandey (1991)
This work

0.2

0.4

0.6

0.8

0
0

0.2

0.4

0.8

0
0

Mode 1
Pandey (1991)
This work

0.6

0.6

0.6

0.4

0.4

0.4

0.2

0.2

0.2

0.4

0.6
x/L

0.8

0
0

0.2

0.4

0.6
x/L

0.6

Mode 2

0.8

0.2

0.4

0.8

x/L

0.8

0
0

0.2

x/L

x/L

0.8

0.6

Mode 3

0.8

0
0

Mode 3

0.2

0.4

0.6

0.8

x/L

Fig. 4. Absolute difference in MCs for the first three modes of the simply supported beam (black dashed line) against the filtered difference in MCs (red
continuous line) for two damage cases: (a)(c) damage located at element 13 (x0 =L 0:6, b=L 0:05, 0.5) as in [19] and (d)(f) x0 =L 0:7, b 0.6 and
0.2. (For interpretation of the references to color in this figure caption, the reader is referred to the web version of this paper.)

and x0 b=2 o L must hold to guarantee the damage to be located inside the beam domain; as a consequence, the contour
plots in Fig. 2 have a triangular shape. Note that all functions i and ik are symmetric with respect to the central point
x0 =L 0:5 as expected from the symmetric boundary conditions of the simply supported beam.
All coefficients i attain their maximum for a damage located at the beam center, i.e., x0 =L 0:5, and spanning the entire
beam length, i.e., b=L-1; on the contrary, 21, 31 and 32 have maxima for different values of the damage parameters. When
the combination of these coefficients is maximum, multiple peaks outside the damage region occur. This is shown in Fig. 3
for three worst-case scenarios: i 1 with x0 =L 0:7 and b/L 0.6 corresponding to 1 maximum; i2 with x0 =L 0:7,
b=L 0:6 that gives a maximum 21; and finally i3 with a damage located at x0 =L 0:5 and b=L 0:57 corresponding to a
maximum 31. For the second mode, for instance, 21 is equal to 0.35 and comparable with 2 0:31; as a consequence,
2 contains a strong contribution of both the first and second MCs which cause the maximum to be located outside the
damage region.

Each i in Fig. 3 is normalized with respect to the damage intensity but their shape is independent of it. This means
that low intensity broad damages (b=L 4 0:4 in Fig. 3) can not be identified through the unfiltered MC difference because of
the multiple peaks outside the damage region. On the contrary, narrow damages b=L o 0:1 produce MC differences with
distinct peaks located in the damage region as found in [19,24]. This is a limitation of the identification techniques based on

the MC difference that has to be taken into account with the proper post-processing of i . This issue can be addressed by
applying the filtering procedure (15).

132

J. Ciambella, F. Vestroni / Journal of Sound and Vibration 340 (2015) 126137

To simulate the experimental data for i and assess the sensitivity of the proposed approach with respect to

discretization and pseudo-experimental errors, a finite element code is used. The resulting MCs and i , post-processed
through Eq. (15), are shown in Fig. 4.
The first case analyzed is the damage scenario, x0 =L 0:6, b/L0.05 with damage intensity 0.5,examined in Pandey's
seminal paper [19] which has been used as a reference in many other works; for consistency, the number of elements in the
finite element model is 20 and the damage is located at element 13 as shown in Fig. 4(ac) with a greyed area. Although
the damage intensity considered in Pandey's work is quite large 0.5, such a severe damage it is used here to demonstrate
the incapability of the unfiltered MC method to predict the damage even in this scenario.
For the first three modes, both the absolute difference of the MCs and the filtered MCs give a clear indication of the
damage position although the unfiltered quantity shows some oscillations outside the damage region (Fig. 4(ac)). The
eigenfrequencies of the simply supported beam are sufficiently sparse and a minimal number of modes is required to filter

the MCs difference: only the first mode for 1 , the first and the second modes for 2 and all three modes for 3 ; the effect of
higher order MC in the series (15) can be neglected.
When the damage is broader, i.e., x0 =L 0:7, b=L 0:6 and 0:2, the absolute values of the MCs difference fails to
provide a clear indication of the damage position with several peaks spread all over the beams (see Fig. 4(df)). However,
once the MCs are filtered according to Eq. (15) and i x is obtained, the damage shape is recovered accurately. Note that the
third MC has a nodal point within the damage region, thus the reconstructed shape in Fig. 4(c) seems to have two separate
peaks associated with only one damage. To overcome the uncertainty that this occurrence could cause, the average sum of
the first three reconstructed damage shape i x can be used; namely
Dx

1 N
b xj
j
N k1 i

(21)

b i x is the ith weighted damage shape function as reconstructed from the experimental data through Eq. (15). There
where
is no need of additional information required to apply Eq. (21) since the computation of the third damage shape already
b i x. Eq. (21) is the damage factor used in [20] for the
requires the knowledge of first two MCs, thus of the first two
unfiltered MC difference.
To further assess the robustness of the proposed procedure, a numerical error with amplitude 5% and 10% is added to the
eigenfrequencies and the eigenvectors, respectively. The results in Fig. 5(ac) show that, in the presence of numerical/
experimental errors, the first two filtered MC differences yet provide a clear localization of the damage, while major
oscillations appear in the unfiltered MCs.

1
0.8

Mode 1
Pandey (1991)
This work

Mode 2

0.8

0.8

0.6

0.6

0.6

0.4

0.4

0.4

0.2

0.2

0.2

0
0

0.2

0.4

0.6

0.8

0
0

0.2

0.4

x/L

0.8

0
0

Mode 1
Pandey (1991)
This work

0.6

0.6

0.6

0.4

0.4

0.4

0.2

0.2

0.2

0.4

0.6
x/L

0.8

0
0

0.2

0.4

0.6
x/L

0.6

Mode 2

0.8

0.2

0.4

0.8

x/L

0.8

0
0

0.2

x/L

1
0.8

0.6

Mode 3

0.8

0
0

Mode 3

0.2

0.4

0.6

0.8

x/L

Fig. 5. Absolute difference in MCs for the first three modes of the simply supported beam (black dashed line) against the filtered difference in MCs (red
continuous line) for the damage case x0 =L 0:7, b=L 0:6 and 0.2 with numerical noise added to simulate the experimental errors (10% on the
eigenvectors and 5% on the eigenfrequencies) (a)(c). The same damage case with the same numerical errors but less elements in the finite element model
(11 instead of 21) (d)(f). (For interpretation of the references to color in this figure caption, the reader is referred to the web version of this paper.)

J. Ciambella, F. Vestroni / Journal of Sound and Vibration 340 (2015) 126137

133

The effect of a reduced number of elements is also assessed and the comparison is shown in Fig. 5(df). A lower number
of measurement points is indeed a common situation when one has to deal with large structures such as bridges (see for
instance [18]). Therefore, the applicability of damage detection techniques to real world scenarios requires a certain degree
of robustness with respect to a low number of measurements. As shown in Fig. 5, despite multiple peaks appearing in
unfiltered MC difference, the proposed filtered quantity still has one clear peak in correspondence with damage for both
modes 1 and 2.
Finally, Fig. 6 shows the applicability of (15) to the case of multiple damages; in particular the figure refers to the case of
1
2
2
1
two damages located at x1
0:5 and
0 =L 0:3 and x0 =L 0:7, with width b =L 0:3 and b =L 0:2 and intensity
2 0:3 respectively. In this case, the unfiltered MC has three distinct peaks, but only one on the actual damage region; on
the contrary, after filtering there are two distinct peaks both of them in correspondence with the damage. The third MC has
a zero which causes the reconstructed damage shape to have two peaks on the right damage; however, this issue is
overcome by applying Eq. (21) and using the combined information of first three damage shapes.

Mode 1
Pandey (1991)
This work

0.8

Mode 2

0.8

0.8

0.6

0.6

0.6

0.4

0.4

0.4

0.2

0.2

0.2

0
0

0.2

0.4

x/L

0.6

0.8

0
0

0.2

0.4

0.6

x/L

0.8

Mode 3

0
0

0.2

0.4

x/L

0.6

0.8

Fig. 6. Absolute difference in MC for the first three modes of the simply supported beam (black dashed line) against the filtered difference in MCs (red
1
2
1
2
continuous line) for two damages (x1
0:3 and x2
0:5). (For interpretation of the references to color in
0 =L 0:3, b =L 0:2,
0 =L 0:7, b =L 0:6,
this figure caption, the reader is referred to the web version of this paper.)

1.0
0.56

0.2

0.8

0.8
0.41

b L

0.24

0.4

0.36
0.32

0.2

0.04

0.16

0.2

0.4 0.6
x0 L

0.8

0.05

0.11

0.28

0.18

0.32

0.18

0.13

0.2

b L

0.36
0.23

0.4

0.8

0.6

0.2

0.4

0.150.26

0.2

0.04

0.4 0.6
x0 L

0.8

1.0

0.05
0.07
0.1
0.02

0.0
0.0

0.2

0.4 0.6
x0 L

0.8

1.0

32

0.8

0.06
0.3

0.07
0.05

0.1

0.13

0.6

0.34
0.13

0.4
0.2

0.02

0.8

1.0

0.23
0.06

0.0 0.03
0.0 0.2

0.1

0.27
0.1

0.17
0.06
0.1

0.07

0.4 0.6
x0 L

0.02

0.2

0.02

0.05

0.2

0.1

1.0

0.23
0.13

0.13

0.16
0.05
0.13
0.18
0.1
0.16
0.08 0.02

0.0
0.0

1.0

0.08

0.24

0.05

0.27
0.09

0.2

0.4 0.6
x0 L
31

0.8

0.41

0.6

0.2

0.21
0.18

0.6

0.2

0.16

1.0

0.46

0.8
b L

0.4

0.0
0.0

1.0

0.0
0.0

0.44

0.11

1.0

0.4

0.22

0.2

0.12

0.0
0.0

0.6

0.33

b L

b L

0.08

0.27

0.8

0.5

0.39

0.28

0.6

21

1.0

b L

1.0

0.17
0.03

0.4 0.6
x0 L

0.8

1.0

Fig. 7. Contour plots with isolines of the functions i and ij and for the first three modes of the cantilever beam (i 1, 2, 3, j o i) against normalized damage
position x0 =L and width b=L.

134

J. Ciambella, F. Vestroni / Journal of Sound and Vibration 340 (2015) 126137

4.2. Cantilever beam

In this section the case of a cantilever beam as already dealt with in [24] is considered. This example is paradigmatic of

the cases where there is no closed-form analytical expression of i but it can be computed numerically. The boundary
conditions of a fixed-free cantilever beam are
v0i
0 0;
x

v0i 0 0;

w0i L 0;

w0i
L 0
x

(22)

and by substituting into (4), the eigenvalues and eigenvectors of the undamaged beam are recovered. A relationship similar
to (20) can still be worked out

~ i xw~ 0i x  i x0 ; bw~ 0i x  ik x0 ; bw~ 0k x

(23)

kai

~ 0i x are the numerical MCs normalized to 1, i x0 ; b : i x0 ; b=i and ik x0 ; bi =k ik x0 ; b maxw 0k x=

where w
0
maxw i x.
Fig. 7 shows the values of i and ik for the first three modes, i.e., i1, 2, 3, k o i. The maximum of is is attained at
different points depending upon the mode number while for the simply supported beam all is had their maximum at
x0 0 =L 0:5 and b=L-1.
1

Fig. 8. Absolute values of MCs differences (thick line) against the actual damage shapes x (thin filled line) for the cantilever beam with three different
damage cases: x0 =L 0:5 and b=L 1 for mode 1, x0 =L 0:3, b=L 0:6 for mode 2 and x0 =L 0:2 and b=L 0:4 for mode 3. In all cases, multiple peaks
outside the damage region appear.

Mode 1

0.8

Pandey (1991)
This work

0.8

0.8

0.6

0.6

0.6

0.4

0.4

0.4

0.2

0.2

0.2

0.2

0.4

0.6

0.8

Mode 2

0.2

0.4

0.6

0.8

1
0.8

0.8

0.6

0.6

0.6

0.4

0.4

0.4

0.2

0.2

0.2

0.4

0.6
x/L

0.4

0.8

0.2

0.4

0.6
x/L

0.6

Mode 2

0.8

0.2

0.2

0.8

x/L

Mode 1
Pandey (1991)
This work

x/L

x/L

Mode 3

0.8

Mode 3

0.2

0.4

0.6

0.8

x/L

Fig. 9. Absolute difference in MCs for the first three modes of the cantilever beam (black dashed line) against the filtered difference in MCs (red continuous
line) for two damage cases: (a)(c) a narrow damage at x0 =L 0:3, b=L 0:05, 0.2 and (d)(f) a broader damage at x0 =L 0:4, b=L 0:4 and 0.2. (For
interpretation of the references to color in this figure caption, the reader is referred to the web version of this paper.)

J. Ciambella, F. Vestroni / Journal of Sound and Vibration 340 (2015) 126137

135

Three worst-case scenarios are shown in Fig. 8 for 1 , 2 and 3 . For the damage positions and widths shown in the
figure, the unfiltered MCs have several peaks outside the damage which could result in an incorrect damage localization.
The effects of filtering through Eq. (15) are shown in Fig. 9 for two damage cases: a narrow damage, i.e., x0 =L 0:3,
b/L 0.05 and 0.2, and a broader damage at x0 =L 0:4 with b/L 0.4 and 0.2. In the former case, both the unfiltered
and filtered MC differences are able to recover the position of the damage accurately for all the modes considered. The
damage area is indeed slightly overestimated by the MC difference, while the filtered quantity is closer to the actual
rectangular shape of the damage (grey area in the figure). In the second damage case, however, the unfiltered quantity
present several peaks outside the damage region making the identification of the damage position prone to errors; on the
contrary, the filtering procedure furnishes a clear indication of both damage position and shape for all the modes.
5. Conclusions
Over the years modal curvatures have been used with different degrees of success as a basic feature of damage detection
techniques applied to aerospace, civil and mechanical structures. While it is widely accepted that MC difference are mainly
located in the damage region, we have shown that several peaks could occur outside the damage region in beam-like
structures independently of the presence of numerical/experimental noise. The amplitude of these peaks has been
quantified and its sensitivity with respect to damage position and width assessed. To the best of our knowledge, such a
sensitivity analysis was yet to be carried out but it has a significant relevancy for assessing the capabilities of existing
damage localization techniques. Even though concentrated damages can still be dealt with, it has been shown that broader
damages cannot be detected by using the absolute difference between damaged and undamaged MCs. This is indeed a
common situation in corroded reinforced concrete beams or composite beams with interplies delaminations.
The perturbative solution of the EulerBernoulli equation has allowed us to quantify and assess the different terms in the
expression of the normalized MC difference responsible of the undesired oscillations outside the damage region. The firstorder solution has been used to obtain a closed form expression of the MCs which demonstrates that, in general, MCs
themselves are not a good feature to localize damages. As such, we have introduced a filtering procedure to reconstruct the
damage shape function i x, that is the product between the actual damage shape x and the ith normalized
undamaged MC.
The introduced filtering procedure has proved to be a suitable and robust mean to detect and localize the occurrence of
single or multiple damages either concentrated or distributed.
Appendix A. Details of the perturbative solution
For the sake of completeness, we report the details of the perturbative solution of the EulerBernoulli equation.
By substituting Eq. (6) into (5), one obtains
ik

kai

d v0k x
4

dx

 i ik v0k x i v0i x 
0

kai

d i x
2

dx

and by a projection over the jth eigenmode v0j x, the previous equation becomes
Z L 4 0
d vk x 0
0
1
ij
vj x dx  i ij Lv0j 2 i ij Lv0i 2
4
0
dx
Z L 2
d i x 0

vj x dx
2
0
dx

(A.1)

(A.2)

where modes orthogonality has been used.

The integral on the left-hand side of (A.2) can be evaluated by scalar multiplying the zeroth order equation (4) by v0j x, i.e.
Z L 4 0
d vk x 0
0
vj xdx k jk Lv0j x2
(A.3)
4
0
dx
while the integral on the right-hand side is
Z
0

L Z L
0
d i x 0
d i x dvj x
vj x 
dx
dx
dx
dx
0
0

L
2
Z L
0
d v0 x
dvj x
i x j 2 dx
 i x

dx
0
dx

d i x
2

dx

v0j xdx

L
0

d v0j x

i x

dx

dx

(A.4)

where the boundary terms are both equal to zero under the assumption that the damage shape x and its derivative are zero
at both ends of the beam.

136

J. Ciambella, F. Vestroni / Journal of Sound and Vibration 340 (2015) 126137

Therefore Eq. (A.2) is rewritten as

ij 0j v0j 2  0i ij v0j 2 1i ij v0i x2 

1
L

L
0

d v0j x

i x

dx

dx

(A.5)

For ij the first-order eigenvalue variation i (Eq. (8)) is obtained, and for i a j, the coefficients ij are recovered (Eq. (9)).
1
For a simply supported beam, i and ij are evaluated in a closed form in terms of the damage parameters x0 and b
defined in Eq. (17)
i
i h
1i 2 3 b3 i3 6 sin ib  2x0 6 sin ib 2x0
2b

6

cos ib  2x0 2 cos ib 2x0  2 cos 2 ix0 ;

(A.6)
3
b
1

ij

 


bi 
cos ix0 cos jx0 cos
2
 

 2
4
2 2
4
h

96i2 j2
4 b3 i j5 i j5 i2 j2


i 2i j 3j 16ij cos
bj i j2
2
 2


bi 
cos ix0 cos jx0 sin
 
2 



4
2 2
4
bj cos bj 3i 2i j j  4 sin bj i4 6i2 j2 j4
2
2




16ij i2 j2 cos ix0 cos jx0 
 


bj 
sin ix0 sin jx0 cos



 
 2 
bi sin bi i4 2i2 j2 3j4 4 cos bi i4 6i2 j2 j4
2
2
 


bj sin ix0 sin jx0 
j sin
2



 
 
bi i2 j2 b cos
bi  3i4 2i2 j2 j4 
16i sin
2 
2

i
4 i4 6i2 j2 j4 sin ix0 sin jx0

bi sin

bj

(A.7)

References
[1] S.W. Doebling, C.R. Farrar, M.B. Prime, A summary review of vibration-based damage identification methods, The Shock and Vibration Digest 30 (1998)
91105.
[2] F. Vestroni, D. Capecchi, Damage detection in beam structures based on frequency measurements, Journal of Engineering Mechanics (2000) 761768.
[3] P.J.S. Cruz, R. Salgado, Performance of vibration-based damage detection methods in bridges, Computer-Aided Civil and Infrastructure Engineering 24
(2009) 6279.
[4] A. Pau, A. Greco, F. Vestroni, Numerical and experimental detection of concentrated damage in a parabolic arch by measured frequency variations,
Journal of Vibration and Control 17 (2010) 605614.
[5] E. Antonacci, A. De Stefano, V. Gattulli, M. Lepidi, E. Matta, Comparative study of vibration-based parametric identification techniques for a threedimensional frame structure, Structural Control and Health Monitoring 19 (2012) 579608.
[6] M.M.F. Yuen, A numerical study of the eigenparameters of a damaged cantilever, Journal of Sound and Vibration 103 (1985) 301310.
[7] X. Ma, M.F.A. Azeez, A.F. Vakakis, A.M.F. Azeez, Non-linear normal modes and non-parametric system identification of non-linear oscillators,
Mechanical Systems and Signal Processing 14 (2000) 3748.
[8] S. Caddemi, A. Morassi, Multi-cracked EulerBernoulli beams: mathematical modeling and exact solutions, International Journal of Solids and Structures
50 (2013) 944956.
[9] G. Kerschen, K. Worden, A.F. Vakakis, J.-C. Golinval, Past, present and future of nonlinear system identification in structural dynamics, Mechanical
Systems and Signal Processing 20 (2006) 505592.
[10] W. Fan, P. Qiao, Vibration-based damage identification methods: a review and comparative study, Structural Health Monitoring 10 (2011) 83.
[11] A. Deraemaeker, E. Reynders, G. De Roeck, Vibration based SHM : comparison of the performance of modal features vs features extracted from spatial
filters under changing environmental conditions, Proceedings of the ISMA, pp. 849863.
[12] M. Dilena, A. Morassi, M. Perin, Dynamic identification of a reinforced concrete damaged bridge, Mechanical Systems and Signal Processing 25 (2011)
29903009.
[13] G.M.L. Gladwell, A. Morassi, Estimating damage in a rod from changes in node positions, Inverse Problems in Science and Engineering 7 (1999) 215233.
[14] J.-T. Kim, Y. Ryu, H. Cho, N. Stubbs, Damage identification in beam-type structures: frequency-based method vs mode-shape-based method, Engineering
Structures 25 (2003) 5767.
[15] E. Parloo, P. Guillame, M. Van Overmeire, Damage assessment using mode shape sensitivities, Mechanical Systems and Signal Processing 17 (2003)
499518.
[16] J. Maeck, M. Abdel Wahab, B. Peeters, G. De Roeck, J. De Visscher, W. De Wilde, J.-M. Ndambi, J. Vantomme, Damage identification in reinforced
concrete structures by dynamic stiffness determination, Engineering Structures 22 (2000) 13391349.
[17] E. Parloo, B. Cauberghe, F. Benedettini, R. Alaggio, P. Guillaume, Sensitivity-based operational mode shape normalisation: application to a bridge,
Mechanical Systems and Signal Processing 19 (2005) 4355.
[18] M. Dilena, A. Morassi, Dynamic testing of a damaged bridge, Mechanical Systems and Signal Processing 25 (2011) 14851507.
[19] A.K. Pandey, M. Biswas, M.M. Samman, Damage detection from changes in curvature mode shapes, Journal of Sound and Vibration 145 (1991) 321332.

J. Ciambella, F. Vestroni / Journal of Sound and Vibration 340 (2015) 126137

137

[20] M. Abdel Wahab, G. De Roeck, Damage detection in bridges using modal curvatures: application to a real damage scenario, Journal of Sound and
Vibration 226 (1999) 217235.
[21] C.P. Ratcliffe, A frequency and curvature based experimental method for locating damage in structures, Journal of Vibration and Acoustics 122 (2000) 324.
[22] C.S. Hamey, W. Lestari, P. Qiao, G. Song, Experimental damage identification of carbon/epoxy composite beams using curvature mode shapes,
Structural Health Monitoring 3 (2004) 333353.
[23] J.F. Unger, A. Teughels, G. De Roeck, Damage detection of a prestressed concrete beam using modal strains, Journal of Structural Engineering 131 (2005)
14561463.
[24] W. Lestari, P. Qiao, S. Hanagud, Curvature mode shape-based damage assessment of carbon/epoxy composite beams, Journal of Intelligent Material
Systems and Structures 18 (2007) 189208.
[25] J. Ciambella, F. Vestroni, S. Vidoli, Damage observability, localization and assessment based on eigenfrequencies and eigenvectors curvatures, Smart
Structures and Systems 8 (2011) 191204.
[26] D. Montalvao, A review of vibration-based structural health monitoring with special emphasis on composite materials, The Shock and Vibration Digest
38 (2006) 295324.
[27] M. Cao, P. Qiao, Novel Laplacian scheme and multiresolution modal curvatures for structural damage identification, Mechanical Systems and Signal
Processing 23 (2009) 12231242.
[28] A. Dixit, S. Hanagud, Single beam analysis of damaged beams verified using a strain energy based damage measure, International Journal of Solids and
Structures 48 (2011) 592602.
[29] M. Cao, W. Xu, W. Ostachowicz, Z. Su, Damage identification for beams in noisy conditions based on Teager energy operator-wavelet transform modal
curvature, Journal of Sound and Vibration 333 (2014) 15431553.