ARTICLE IN PRESS

JOURNAL OF
SOUND AND
VIBRATION
Journal of Sound and Vibration 275 (2004) 577590
www.elsevier.com/locate/jsvi

Damage identication in structures using the

time-domain response
S. Choia,*, N. Stubbsb
a

Structural Systems and Site Evaluation Department, Korea Institute of Nuclear Safety,
Yoosung PO Box 114, Daejeon 305-600, South Korea
b
Department of Civil Engineering, Texas A&M University, College Station, TX 77843, USA
Received 19 August 2002; accepted 26 June 2003

Abstract
A methodology to locate and size damage in a structure using the time-domain response is presented in
this paper. The measured response in the time domain is spatially expanded over the structure and the mean
strain energy for a specied time interval is obtained for each element of the structure. The mean strain
energy for each element is, in turn, used to build a damage index that represents the ratio of the stiffness
parameter of the pre-damaged to the post-damaged structure. The damage indices are used to identify
possible locations and corresponding severities of damage in the structure. The validity of the methodology
is demonstrated using data from a numerical example of a beam structure.
r 2003 Elsevier Ltd. All rights reserved.

1. Introduction
As concerns about the soundness of the civil engineering structures grow, the need to develop
effective and practical structural health monitoring (SHM) systems is emerging as a prominent
problem in structural engineering. SHM system includes a synthesis of experimental data
gathering techniques and non-destructive damage evaluation (NDE) schemes. To date, numerous
NDE methods have been proposed. The theoretical bases of these methods include neural
networks [1], pattern recognition [2], sensitivity analysis [3,4], fuzzy sets [5], and system
identication [6,7]. Recently, methods that attempt to simultaneously assess the condition of
the whole structure, the so-called global NDE methods, have been gaining acceptance in
*Corresponding author. Tel.: +82-42-868-0666; fax: +82-42-868-0523.
E-mail addresses: [email protected] (S. Choi), [email protected] (N. Stubbs).
0022-460X/$ - see front matter r 2003 Elsevier Ltd. All rights reserved.
doi:10.1016/j.jsv.2003.06.010

ARTICLE IN PRESS
578

S. Choi, N. Stubbs / Journal of Sound and Vibration 275 (2004) 577590

the engineering community. The basic idea behind these global methods is that changes in the
physical properties of a system alter the response characteristics of the structure. A recent round
robin study involving ve established NDE methods reveals that the damage index method [8]
performed best using simulated data and data from a real structure [9].
The damage index method originated from a sensitivity approach that relates changes in modal
responses, specically resonant frequencies, to changes in the mass, damping, and stiffness
of a structure. Based on work done by Cawley and Adams [10], Stubbs [11], and Stubbs and
Rials [12] proposed a method that relates changes in the resonant frequencies to changes
in element stiffness. To reduce the computational effort involved in the proposed scheme,
Stubbs [13] developed an NDE theory using linear inverse method. Stubbs and Osegueda
[3,4] further developed the method for applications to beams and other structures. To overcome the difculties of the previous works, which arise when the number of vibrational modes is
much less than the number of damage parameters to be determined, Stubbs et al. [8] developed the
so-called damage index method. This method has been corroborated using (1) numerically
simulated data for various structural types and classes [14,15], (2) experimental modal data
generated in a laboratory environment [16], and (3) eld data measured on bridge structures
[9,17].
In recent years, most global NDE methods, including the damage index method, utilize
modal data (i.e., resonant frequencies and modeshapes) to predict possible damage locations
and estimate the severity of the damage. The modal data can be extracted using
established experimental modal analysis techniques (inputoutput methods or output only
methods) to estimate the modal parameters (i.e., resonant frequencies, modal damping,
and mode shapes) [18]. However, the inputoutput modal parameter extraction method
involves averaging and curve-tting procedures which introduce additional uncertainties
and measurement errors. Furthermore output-only extraction methods are computationally intensive. One potential solution to this problem is to investigate other types of response
measures (e.g., static response [19], time-domain response [20], etc.) that circumvent these
difculties.
The objective of this paper is to develop a time-domain-based NDE methodology that obviates
the computational demands, complexity, and subjectivity associated with current modal
parameter extraction methods. In the proposed method, a mean strain energy measure which
uses response data for a specied time period is utilized to formulate a damage index for an
element in a structure. The damage indices, which represent the ratio of pre- and post-damaged
stiffness of the elements, are utilized to identify the locations and corresponding severities of the
possible damage locations. The validity of the methodology is demonstrated using numerically
generated data from a continuous beam structure.

2. Theory
2.1. New damage ratio using the time-domain response
Suppose that dynamic responses of a structure are sampled at n locations at time interval Dt
from ti t1 to ti tNT ; where NT is the number of sampling points. Then, the sampled dynamic

ARTICLE IN PRESS
S. Choi, N. Stubbs / Journal of Sound and Vibration 275 (2004) 577590

579

responses for the n sensor locations may be expressed in matrix form as follows:
Locations - 1
2

w1;1
6
6 ^
6
W 6
6 wi;1
6
4 ^
wNT;1

j
?

w1;j

&

wi;j

wNT;j

n
w1;n

Time Step
3

1
7
c
^ 7
7
?
wi;n 7
7 i
7
&
^ 5
? wNT;n NT
?

In the matrix, the jth column vector represents dynamic displacements at the jth location, and the
ith row vector represents the dynamic displacements at the ith time step. Using current
technology, the discretized displacements can be obtained by integrating the acceleration records
obtained from accelerometers [21]. Thus, each row vector describes the instantaneous deformed
conguration of the structure at a certain instant of time. Consequently, there are NT response
measurement sets, if a single row vector is regarded as one measurement set.
Since damage in a structure causes changes in the dynamic characteristics of the system such as
frequencies and mode shapes, the displacement congurations of the pre-damaged and the postdamaged structures at the same sampling time will be different. Thus, the deformed conguration
of the pre-damaged structure measured at a certain time should not be compared to the deformed
conguration of the post-damaged structure at the same time. To overcome this difculty, the
mean strain energy of the structure over a sampling period is used here instead of the direct
comparison between strain energies of pre- and post-damaged structures at the same time. The
matrix expression presented in Eq. (1) may be rewritten as
W V1 ; y; Vi ; y; VNT T ;

where a vector, Vi V ti ; represents a displacement conguration of the structure at time ti :

Considering a system with NE elements, the corresponding strain energy of the structure at a time
ti can be expressed as
s Ui

12 ViT KVi ;

where the subscript s refers to the structure. The mean strain energy for a specied time interval
between ta and tb b > a; s U% t%; may be dened as
% t% Es Ui 
sU

b
X
1
V T KVi :
2b  a ia i

Similarly, the average mean strain energy for the jth element in a structure may be given by
% t%
jU

Ej Ui 

b
X
1
V T Kj Vi ;
2b  a ia i

where Kj kj Cj ; kj is the stiffness of element j; and Cj is the geometric portion of the

contribution of the jth element to the system stiffness matrix. The ratio of the mean strain energy

ARTICLE IN PRESS
580

S. Choi, N. Stubbs / Journal of Sound and Vibration 275 (2004) 577590

for the jth element to the system mean strain energy is given by
P
% t% kj bia ViT Cj Vi
jU
Fj
Pb
;
T
% t%
sU
ia Vi KVi

where j U% t% represents the average strain energy stored in the jth element. Fj will be denoted as the
fractional mean strain energy of element j: Similarly, for the damaged structure:
P
% t%
kj bia ViT Cj Vi
jU

:
7
Fj  Pb
T  
% t%
sU
ia Vi K Vi
The pre-damaged and the post-damaged fractional mean strain energy for jth element are
related by
8
Fj Fj dFj ;
where dFj is related to the change in the fractional mean strain energy of the jth element resulting
from the damage. The quantity dFj can be obtained from the rst order expansion [22]:
dFj E  Fj aj ;

where the fractional change in stiffness, aj ; is given by

dkj kj  kj
aj

:
10
kj
kj
Dene a damage index for the jth element as bj kj =kj : Then substituting Eqs. (6), (7), (9), and
(10) into Eq. (8) and simplifying yields the damage index for the element j as
!
kj
1 fj
1 ;
11
bj 
2 fj
kj
where the quantity fj is given by

Pb

fj Pia
b

ia

and

ViT Cj Vi
ViT KVi

Pb
V T Cj Vi

fj Pbia i
:
T  
ia Vi K Vi

12

13

Note that the damage index obtained using Eq. (11) is most susceptible to measurement and
numerical errors when both numerator and denominator are close to zero. This phenomenon, for
example, might be observed for elements near supports when the element size and the
displacements are small. In such cases, localization errors may result. To avoid this problem, the
domain of interest in the problem is shifted by adding unity to the denominator and numerator of
Eq. (11) [22]. The resulting non-singular damage index is then given by the equation
!
PNT T
PNT T
 PNT V T KV 
V
C
V
V
KV
j i
i
i
i
i
i1 i
i1
:
14
bj EPNT
P
Pi1
NT
NT
TC V
T KV
T KV 
V
V
V
j i
i
i
i1 i
i1 i
i1 i

ARTICLE IN PRESS
S. Choi, N. Stubbs / Journal of Sound and Vibration 275 (2004) 577590

581

Note that the damage index shown in Eq. (14) is the ratio of the effective stiffness of an element
of the pre-damaged state to the post-damaged state.
2.2. Damage localization using the new damage ratio
Possible locations of damage in a structure can be identied by utilizing classication
algorithms with the damage index given by Eq. (14) taken as the feature vector. As presented in a
prior study [2], classication of an element as damaged or not damaged can be made on the basis
of such schemes as (1) Bayes rule (from which the well-known Linear and Quadratic
Discriminant Analysis are derived) [23]; (2) nearest distance [24]; and (c) hypothesis testing
[25]. In this study, hypothesis testing is used for the classication of an element as being damaged
or not damaged. In hypothesis testing, the alternate hypothesis (H1 ) and null hypothesis (H0 ) are
dened as H0 : element j of the structure is not damaged, H1 : element j of the structure is damaged.
To test the hypotheses, the damage indices shown in Eq. (14) are standardized using the
equation
bij  mb
:
15
zij
sb
Here, it is assumed that the damage index, bij ; is a random variable and the collection of the
damage indices are distributed normally. Thus, a typical probability density function of the
standardized damage indices for elements can be depicted as shown in Fig. 1. The one-tailed test
to decide on the existence of damage in an element may be restated as (1) choose H0 if zj ozZ ; (2)
choose H1 if zj XzZ ; where the quantity zb is the standardized damage index, the quantity zZ is a
threshold value of zb ; and Z represents the level of signicance of the test (see Fig. 1). The decisionmaking criterion for assigning the location of damage is thus established using elements of
statistical decision-making. One can choose a greater threshold value to have more condence in
the identied damage locations. Note that a typical value for the level of signicance in damage
localization is 0.05 which corresponds to a z score of z0:05 1:645:
2.3. Severity estimation using the new damage ratio
Once the possible locations of damage are isolated, corresponding damage severities can be
obtained using the corresponding damage indices to localized damage locations. Since the damage
index is the ratio of pre-damage stiffness to post-damage stiffness, the severity of damage for jth

fz (z)

PDF of standardized

Fig. 1. Assumed probability density functions of standardized damage index.

ARTICLE IN PRESS
S. Choi, N. Stubbs / Journal of Sound and Vibration 275 (2004) 577590

582

element (i.e., the fractional loss in stiffness) may be expressed as

aj

kj  kj
kj

1
 1:
bj

16

Note that the severity of damage obtained using Eq. (16) represents the effective stiffness loss to
undamaged stiffness for a specic element j of the structure.

3. Verication of methodology using simulated damage

The feasibility and performance of the proposed NDE algorithm is examined via a numerical
example of a continuous beam structure. The example structure, shown in Fig. 2, is a three-span
plate girder, and consists of two cross-sectional properties designated here as thick- and thinanged girder elements.
The sectional and material properties of the members of the structure are summarized in
Table 1. Two-node cubic beam elements are used to model the example structure. The model has
60 elements and 61 nodes, as shown in Fig. 2. The structure is subjected to 15 damage scenarios.
The locations and corresponding magnitudes of the damage simulated for each damage scenarios
are summarized in Table 2. The damage is numerically simulated by reducing the elastic modulus
of the appropriate elements. Forced vibration analysis was performed to compute time responses
using a commercial FORTRAN code [26]. An impulse load is used to excite the structure at Node
31 in the vertical direction. The impact loading is shown in Fig. 3 and the location of the
excitation is shown in Fig. 2. A direct integration method, the Newmark-b method [27], was used
to generate the time response data. A magnitude of 0.1% viscous damping, which is typical for
steel structures [28], was assumed for the structure. The displacement response data are assumed
to be measured at 61 locations (n 61) which correspond to the node points in Fig. 2. In real
world applications using currently available technology, such a dense arrangement is not
practical; however, the purpose of this paper is to establish the feasibility of the approach. The
impact of using fewer sensors per span will be addressed in subsequent studies. Note that
responses of the structure were sampled at each node for 2 s in time intervals of 0:002 s
(Dt 0:002) after the loading had been applied. Thus, a total of 1000 discrete displacement
measurements (NT 1000) were sampled for each node.
Node Number
(Sensor Locations)
21

Thick-Flanged Girder

41

61

3031

12

Element Number
[email protected] = 39.62m

60

Thin-Flanged Girder (Typical)

[email protected] = 49.68m

[email protected] = 39.62m

Fig. 2. Schematic of the nite element model of the continuous beam.

ARTICLE IN PRESS
S. Choi, N. Stubbs / Journal of Sound and Vibration 275 (2004) 577590

583

Table 1
Sectional and material properties of the beam structure

Area (cm2 )
Second moment of inertia (m4 )
Depth of the girder (cm)
Flange thickness (cm)
Flange width (cm)
Web thickness (cm)
Elastic modulus (GPa)
Mess density (kg=m3 )

Thick-anged girder

Thin-anged girder

1103
0.2197
305
6.67
61
0.95
200
7850

697
0.1193
305
3.81
53.34
0.95
200
7850

Table 2
Simulated damage locations and severities
Damage scenario

Elements
damaged

Corresponding
severity (%)

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15

5
32
54
20
56
15
12
7
26
3
33
42
50
53
6

6.0
16.0
9.0
3.0
7.0
7.0
16.0
3.0
8.0
4.0
3.0
7.0
3.0
7.0
2.0

10
37
21
31
42
11
34
15
9
35

47
8
4
16
40

9.0
7.0
5.0
4.0
2.0
5.0
3.0
12.0
8.0
5.0

12.0
4.0
5.0
13.0
8.0

p(t)
4450 N

0.02

t (sec)

Fig. 3. Dynamic loading used to excite the continuous beam.

3.1. Measurement noise

In eld application, it is expected that there would be some deviations due to measurement
noise. In this example, the measurement noise is simulated by adding a series of random numbers

ARTICLE IN PRESS
584

S. Choi, N. Stubbs / Journal of Sound and Vibration 275 (2004) 577590

to the obtained mean strain energies for elements in the example structure. The random numbers
are generated from a uniform distribution on the interval 1; 1: The degree of noise is
determined by a noise/signal (NS) ratio that is a ratio of the random-number series to the
amplitude of the mean strain energy [29]. In present applications, the effect of different level of
noise on damage identication is investigated by applying 1% and 2% NS ratios.
3.2. Damage localization and severity estimation results
Using the damage index expression, presented in Eq. (14), the determination of the locations of
potential damage in the structure is implemented using the following steps. First, the damage
indices for each element are calculated using Eq. (14). Second, the obtained damage indices are
standardized using Eq. (15). Third, the presence of damage in element j is determined according to
the pre-assigned classication rules: (1) the element is damaged if zj X2:0; (2) the element is not
damaged if zj o2:0: Note that the value of the damage indicator, 2.0, corresponds to a 0.02 level of
signicance test for the presence of damage.
The severity of damage is estimated using Eq. (16). The severity of damage for possible damage
locations was estimated as follows. First, the possible damage elements are identied using the
hypothesis testing algorithm. Second, the severity of damage is estimated using the damage indices
for the possible damaged element using Eq. (16).
3.3. Results
The damage localization results are shown in Figs. 46. In all of the gures, note that the
inicted locations of damage are indicated by the tilted arrows near the horizontal axis. The
vertical axis is in the non-dimensional units of the standardized damage ratios for that particular
location. The percentage of false positives predictions (Type I error) and the percentage of false
negatives predictions (Type II error) are used to evaluate the performance of the methodology. A
false positive means that damage is reported where no damage exists and a false negative means
that damage is not reported where damage exists. The percentage of false positives is calculated by
dividing the number of false positive predictions by the number of undamaged elements, and the
percentage of false negatives is calculated by dividing the number of false negative predictions by
the number of damaged elements. The percent of false positives may reect the quality of the
measured data as well as the effectiveness of the damage localization algorithm while the
percentage of false negatives may measure the sensitivity of measures to damage and the ability of
the classication algorithm to correctly identify the damage location. The resulting percentage of
false positives and negatives are summarized in Table 3.
For damage cases with a single damage location that correspond to damage cases 15, the
proposed methodology successfully identies all simulated damage locations (zero false negatives)
with noise free data (see Fig. 4). Few false positives were observed but all the false positive
predictions arise in the neighborhood of the simulated damage locations. When noise added,
the methodology identied all damage locations except damage case 4 where the magnitude of the
simulated damage was relatively small. As the NS ratio increased, it was observed that the
standardized value for the damage index at a damage location was decreased.

ARTICLE IN PRESS
S. Choi, N. Stubbs / Journal of Sound and Vibration 275 (2004) 577590
8
Damage Indicator, Zj

Damage Indicator, Zj

8
6
4
2
0
-2
1

2
0
-2
1

Damage Indicator, Zj

Damage Indicator, Zj

6
4
2
0
-2
6 11 16 21 26 31 36 41 46 51 56

Damage Indicator, Zj

Element Number

Element Number
8
6
4
2
0
-2
1

6 11 16 21 26 31 36 41 46 51 56

(d)

Element Number

8
6
4
2
0
-2
1

(e)

6 11 16 21 26 31 36 41 46 51 56

(b)

6 11 16 21 26 31 36 41 46 51 56
Element Number

(a)

(c)

585

6 11 16 21 26 31 36 41 46 51 56
Element Number

Fig. 4. Damage localization results for damage Cases 15: (a) damage case 1, (b) damage case 2, (c) damage case 3,
(d) damage case 4, (e) damage case 5. location of damage;
; noise free; - - - -, 1% NS; , 2% NS.

In the damage localization results for damage cases with multiple damage locations, damage
cases 615, the proposed localization methodology performed quite satisfactorily with noise free
data (see Figs. 5 and 6). In only two damage scenarios, damage case 13 and 15 in which the
structure was damaged at 3 locations, the standardized value for the damage index was between 1
and 2. However, all three damage locations in those two damage cases can be inferred from Fig. 6.
From a review of the damage magnitudes inicted at the various locations and presented in
Table 2, one explanation for the lower value of the standardized damage index in the latter cases is
the possible masking of the smallest damage magnitudes at location 50 in damage case 13 and
location 6 in damage case 15. When noise was added, it was observed that the percentage of false
negatives was increased as the NS ratio increased. However, even with the severest simulated noise
level when the NS ratio was 2%, most of damage locations can be inferred from Figs. 5 and 6. It
can be seen in the gures that, with a lower signicance level of the test (less condence in the
presence of damage), the detectability of the method can be improved dramatically. For example,
when the NS ratio is 2%; with the threshold value of 1.5, the number of false negatives can be

ARTICLE IN PRESS
S. Choi, N. Stubbs / Journal of Sound and Vibration 275 (2004) 577590

586

8
Damage Indicator, Zj

Damage Indicator, Zj

8
6
4
2
0
-2
1

0
-2
1

Damage Indicator, Zj

Damage Indicator, Zj

6
4
2
0
-2
6 11 16 21 26 31 36 41 46 51 56
Element Number

Damage Indicator, Zj

(c)

(e)

6 11 16 21 26 31 36 41 46 51 56
Element Number

(b)

6 11 16 21 26 31 36 41 46 51 56
Element Number

(a)

8
6
4
2
0
-2
1

6 11 16 21 26 31 36 41 46 51 56

(d)

Element Number

8
6
4
2
0
-2
1

6 11 16 21 26 31 36 41 46 51 56
Element Number

Fig. 5. Damage localization results for damage Cases 610: (a) damage case 6, (b) damage case 7, (c) damage case 8,
(d) damage case 9, (e) damage case 10. location of damage;
; noise free; - - - -, 1% NS; , 2% NS.

decreased to three from nine in all damage cases with multiple damage locations (Figs. 5 and 6).
From Table 3, the percentage of false positives seemed not to be inuenced signicantly by the
increased number of damage locations and the noise level. However, it can be inferred from
the gures that the number of false positives will be increased with a lower signicance level of the
test.
Another way to evaluate the performance of the proposed method is as follows. In all 15
damage cases a total of 30 damage sites were inicted. Without noise, the method unambiguously
located 28 of these 30 cases. This result corresponds to an overall success rate of 93%. With
additional screening of the data or lowering the signicance level of the test, the success rate for
the sample used here could be as high as 100%. When noise was added, the method identied 26
and 21 in 30 simulated damage locations, which corresponds a success rate of 87% and 70%,
respectively.
The severity estimation results using the proposed methodology are presented in Table 4. For
all damage cases, the proposed methodology consistently yields lower damage severity estimates

ARTICLE IN PRESS

Damage Indicator, Zj

Damage Indicator, Zj

S. Choi, N. Stubbs / Journal of Sound and Vibration 275 (2004) 577590

6
4
2
0
-2
1

4
2
0
-2
1

Damage Indicator, Zj

Damage Indicator, Zj

6
4
2
0
-2
6 11 16 21 26 31 36 41 46 51 56

Damage Indicator, Zj

Element Number

Element Number
8
6
4
2
0
-2
1

6 11 16 21 26 31 36 41 46 51 56

(d)

Element Number

8
6
4
2
0
-2
1

(e)

6 11 16 21 26 31 36 41 46 51 56

(b)

(c)

6 11 16 21 26 31 36 41 46 51 56
Element Number

(a)

587

6 11 16 21 26 31 36 41 46 51 56
Element Number

Fig. 6. Damage localization results for damage Cases 1115: (a) damage case 11, (b) damage case 12, (c) damage case
13, (d) damage case 14, (e) damage case 15. location of damage;
; noise free; - - - -, 1% NS; , 2% NS.

than the simulated values. The fact that damage is smeared into the neighboring elements, as
shown in Figs. 46, may explain this systematic error. In the table, it was also observed that the
proposed methodology gives a reasonable agreement between damage magnitudes using the
noise-free and the noise-polluted data.

4. Summary and conclusion

In this paper, a damage localization and severity estimation methodology using time-domain
response data was presented. The motivation for the approach was to investigate the possibility of
obviating certain drawbacks associated with commonly used modal extraction parameter
methods. The time-domain displacement response measurements were assumed to be measured at
a nite number of locations in a structure and the mean strain energy over a specied time interval
was obtained for each and every element of the structure. The mean strain energy for the elements

ARTICLE IN PRESS
588

S. Choi, N. Stubbs / Journal of Sound and Vibration 275 (2004) 577590

Table 3
Number of false positives and false negatives
Damage case

False positives
No noise

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
S
a
b

2
1
1
0
2
0
0
0
0
0
1
1
0
0
0

(3 )
(2)
(2)
(0)
(3)
(0)
(0)
(0)
(0)
(0)
(2)
(2)
(0)
(0)
(0)

8 (1)

False negatives
1% noise
1
1
0
1
1
0
0
1
0
2
0
1
0
0
1

(2 )
(2)
(0)
(2)
(2)
(0)
(0)
(2)
(0)
(3)
(0)
(2)
(0)
(0)
(2)

9 (1)

2% noise

No noise

(0 )
(2)
(2)
(2)
(0)
(2)
(0)
(3)
(2)
(2)
(0)
(2)
(0)
(0)
(2)

0
0
0
0
0
0
0
0
0
0
0
0
1
0
1

(0 )
(0)
(0)
(0)
(0)
(0)
(0)
(0)
(0)
(0)
(0)
(0)
(33)
(0)
(33)

0
0
0
1
0
0
0
0
0
0
1
0
1
0
1

10 (1)

(7)

4 (13)

0
1
1
1
0
1
0
2
1
1
0
1
0
0
1

1% noise

(0 )
(0)
(0)
(100)
(0)
(0)
(0)
(0)
(0)
(0)
(33)
(0)
(33)
(0)
(33)

2% noise
0
0
0
1
0
1
1
1
0
1
0
1
2
0
1

(0b)
(0)
(0)
(100)
(0)
(50)
(50)
(50)
(0)
(50)
(0)
(67)
(33)
(0)
(33)

9 (30)

Percentage of false positives.

Percentage of false negatives.

was, in turn, used to build an element damage index dened as the ratio of the stiffness parameter
of the pre-damaged to the post-damaged elements. The standardized damage indices were then
used as feature vectors in a classication scheme to identify damage. Depending upon the value of
the standardized damage index for a given element, the element was classied as either damaged
or undamaged. The classication scheme used here was based on the statistical decision technique
of hypothesis testing.
The feasibility of the methodology was demonstrated using simulated data from a continuous
beam structure. The performance of the proposed methodology was evaluated in terms of the
number of damage locations false-positive predictions, the number of false-negative predictions,
and the accuracy of the damage severity predictions. A total of 15 damage cases were simulated.
In these cases a total of thirty damage locations were simulated. Also, the robustness of the
method to noise was investigated by adding the NS ratio of 1% and 2% to the noise-free data.
With the same signicance level of the test, the method correctly identied 93%, 87%, and 70% of
the damage events using the noise-free, the 1% NS ratio, and the 2% NS ratio data, respectively.
From the numerical study, the following conclusions are drawn: (1) the time-domain response
data may be used directly to localize and size damage in a structure; (2) the numerical simulation
of a continuous beam reveals that the proposed methodology can identify single and multiple
damage locations consistently and accurately even using the data with simulated noise; (3) the
false negatives can be reduced using the lowered signicance level of test for damage localization
at the expense of increasing the number of false positives; and (4) the proposed methodology
consistently produces lower damage severity estimations.

ARTICLE IN PRESS
S. Choi, N. Stubbs / Journal of Sound and Vibration 275 (2004) 577590

589

Table 4
Severity estimation results
Damage case

1
2
3
4
5
6
7
8
9
10
11

12

13

14

15

Element number

5
32
54
20
56
10
15
12
37
7
21
26
31
3
42
11
33
47
8
34
42
4
15
50
9
16
53
6
35
40

Simulated severity %

6.0
16.0
9.0
3.0
7.0
9.0
7.0
16.0
7.0
3.0
5.0
8.0
4.0
4.0
2.0
5.0
3.0
12.0
4.0
3.0
7.0
5.0
12.0
3.0
8.0
13.0
7.0
2.0
5.0
8.0

Estimated severity (%)

No noise

1% noise

2% noise

3.8
9.4
5.4
1.4
4.5
4.8
3.8
9.9
3.9
1.4
2.7
4.4
1.7
2.5
1.1
2.8
1.9
7.5
1.9
1.8
4.2
3.8
7.3

3.7
7.5
3.5

2.9
4.6

2.9
9.6
4.5

4.6
4.6
3.7
9.5
3.9
1.2
2.6
4.3
0.8
2.6
1.5

1.8
7.6
2.7
2.0
3.2
3.6
6.9

3.8
7.2
3.3

3.6
4.2

4.5
8.3
5.1

5.1
3.9

10.4

3.3
4.4
2.0
4.3
1.5
4.0
3.2
7.3

1.9
3.7

8.4

4.3
7.6
3.5

4.1
3.9

References
[1] S.F. Masri, A.G. Chassiakos, T.K. Caughey, Identication of nonlinear dynamic systems using neural networks,
Journal of Applied Mechanics 60 (1993) 123133.
[2] N. Stubbs, G. Garcia, Application of pattern recognition to damage localization, Microcomputers in Civil
Engineering 11 (1996) 395409.
[3] N. Stubbs, R. Osegueda, Global non-destructive damage evaluation in solids, The International Journal of
Analytical and Experimental Modal Analysis 5 (2) (1990) 6779.
[4] N. Stubbs, R. Osegueda, Global damage detection in solidsexperimental verication, The International Journal
of Analytical and Experimental Modal Analysis 5 (2) (1990) 8197.
[5] J.T.P. Yao, Damage assessment of existing structures, Journal of Engineering Mechanics Division 106 (4) (1980)
785799.

ARTICLE IN PRESS
590

S. Choi, N. Stubbs / Journal of Sound and Vibration 275 (2004) 577590

[6] H.G. Natke, Updating computational models in the frequency domain based on measured data: a survey,
Probabilistic Engineering Mechanics 8 (1988) 2835.
[7] J.E. Mottershead, M.I. Friswell, Model updating in structural dynamics: a survey, Journal of Sound and Vibration
167 (1993) 347375.
[8] N. Stubbs, J.T. Kim, K.G. Topole, An efcient and robust algorithm for damage localization in offshore
platforms, ASCE 10th Structures Congress92, 1992, pp. 543546.
[9] C. Farrar, D. Jauregui, Damage detection algorithms applied to experimental and numerical modal data from the
I-40 bridge, Technical Report LA-13074-MS, Los Alamos National Laboratory, 1996.
[10] P. Cawley, R.D. Adams, The location of defects in structures from measurements of natural frequencies, Journal of
Strain Analysis 14 (2) (1979) 4957.
[11] N. Stubbs, Non-destructive evaluation of damage in periodic structures, Proceedings of the Third International
Conference on Space Structures, 1984, pp. 332337.
[12] N. Stubbs, D. Rials, Non-destructive evaluation of damage in multi-story structure, Structural Safety and
Reliability 2 (1985) 517522.
[13] N. Stubbs, A general theory of non-destructive damage detection in structures, in: H.H.H. Leipholz (Ed.),
Structural Control, 1985, pp. 694713.
[14] S. Choi, N. Stubbs, Nondestructive damage detection algorithms for 2d plates, Smart Structures and Materials
1997: Smart Systems for Bridges, Structures, and Highways, SPIE Proceedings, Vol. 3043, 1997, pp. 193204.
[15] J.T. Kim, N. Stubbs, Damage detection in offshore jacket structures from limited modal information, International
Journal of Offshore and Polar Engineering 5 (1) (1995) 5866.
[16] J.T. Kim, N. Stubbs, Model-uncertainty impact and damage-detection accuracy in plate girder, Journal of
Structural Engineering 121 (10) (1995) 14091417.
[17] S. Park, N. Stubbs, R. Bolton, S. Choi, C. Sikorsky, Field verication of the damage index method in a concrete
box-girder bridge via visual inspection, Computer-Aided Civil and Infrastructure Engineering 16 (2001) 5870.
[18] D.J. Ewins, Modal Testing: Theory and Practice, Research Studies Press, Hertfordshire, England, 1986.
[19] M. Sanayei, O. Onipede, Damage assessment of structures using static test data, American Institute of Aeronautics
and Astronautics Journal 29 (7) (1991) 11741179.
[20] M.S. Agbabian, S.F. Masri, R.K. Miller, T.K. Caughey, A system identication approach to the detection of
structural changes, Journal of the Engineering Mechanics 117 (2) (1990) 370390.
[21] B.C. Faulkner, F.W. Barton, T.T. Baber, W.T. Mckeel Jr., Determination of bridge response using acceleration
data, Technical Report VTRC-97-R5, Virginia Transportation Research Council, 1996.
[22] S. Park, Development of a Methodology to Continuously Monitor the Safety of Complex Structures, PhD
Dissertation, Texas A&M University, College Station, TX, 1997.
[23] J.D. Gibson, J.L. Melsa, Introduction to Nonparametric Detection with Applications, Academic Press, New York,
1975.
[24] M. Nadler, E.P. Smith, Pattern Recognition Engineering, Wiley, New York, 1993.
[25] R.L. Ott, An Introduction to Statistical Methods and Data Analysis, Wadsworth, Belmont, CA, 1993.
[26] ABAQUS, Version 5.4 Users Manual, Hibbitt, Karlsson & Sorensen Inc., Providence, RI, 1994.
[27] J.N. Reddy, An Introduction to the Finite Element Method, McGraw-Hill, New York, 1993.
[28] R.R. Craig, Structural DynamicsAn Introduction to Computer Methods, Wiley, New York, 1981.
[29] W.-X. Ren, G.D. Roeck, Structural damage identication using modal data I: simulation verication, Journal of
Structural Engineering 128 (1) (2002) 8795.