You are on page 1of 11

Engineering Structures 25 (2003) 57–67

www.elsevier.com/locate/engstruct

Damage identification in beam-type structures: frequency-based


method vs mode-shape-based method
Jeong-Tae Kim a,∗, Yeon-Sun Ryu a, Hyun-Man Cho a, Norris Stubbs b
a
Department of Ocean Engineering, Pukyong National University, Nam-gu, Pusan 608-737, South Korea
b
Department of Civil Engineering, Texas A&M University, USA

Received 25 March 2001; received in revised form 22 July 2002; accepted 26 July 2002

Abstract

This paper presents a methodology to nondestructively locate and estimate the size of damage in structures for which a few
natural frequencies or a few mode shapes are available. First, a frequency-based damage detection (FBDD) method is outlined. A
damage-localization algorithm to locate damage from changes in natural frequencies and a damage-sizing algorithm to estimate
crack-size from natural frequency perturbation are formulated. Next, a mode-shape-based damage detection (MBDD) method is
outlined. A damage index algorithm to localize and estimate the severity of damage from monitoring changes in modal strain energy
is formulated. The FBDD method and the MBDD method are evaluated for several damage scenarios by locating and sizing damage
in numerically simulated prestressed concrete beams for which two natural frequencies and mode shapes are generated from finite
element models. The result of the analyses indicates that the FBDD method and the MBDD method correctly localize the damage
and accurately estimate the sizes of the cracks simulated in the test beam.
 2002 Elsevier Science Ltd. All rights reserved.

1. Introduction localize damage using changes in natural frequencies


[4,13–16]). The most appealing feature associated with
Nondestructive damage detection (NDD) is an using frequencies is that natural frequencies are rela-
important subproblem of damage assessment and should tively simple to measure. However, the feasibility of
form the basis of any decision to repair, rehabilitate, or using frequency changes for the purpose of damage
replace a structure. For critical structural systems such localization is limited for at least two reasons. First, sig-
as aircraft, bridges, and offshore platforms, an accurate nificant damage may cause very small changes in natural
and reliable NDD capability of the structural systems is frequencies, particularly for larger structures, and these
essential, since damage that is not detected and not changes may go undetected due to measurement or pro-
repaired may lead to catastrophic structural failure. cessing errors. Next, variations in the mass of the struc-
During the past two decades, a significant amount of ture or measurement temperatures may introduce uncer-
research has been conducted in the area of NDD via tainties in the measured frequency changes.
changes in the modal response of a structure (see e.g., In an effort to overcome these difficulties, research
[1–10]). Also, alternative research efforts using the efforts have focused on using changes in mode shapes
wavelet analysis have been made to detect damage in [10,17–20]. The appealing feature is that changes in
beam-type structures for which the effect of a small mode shapes are much more sensitive to local damage
crack may not be evident from changes to eigenfrequ- when compared to changes in natural frequencies. How-
encies of the structures [11,12]. ever, using mode shapes also has some drawbacks. First,
Many researchers have attempted to detect and damage is a local phenomenon and may not significantly
influence mode shapes of the lower modes that are usu-
ally measured from vibration tests of large structures.

Corresponding author. Tel.: +82 51 620 6227; fax: +82 51 628 Second, extracted mode shapes are affected by environ-
8146. mental noises from such sources as ambient loads or
E-mail address: idis@pknu.ac.kr (J.-T. Kim). inconsistent sensor positions. Third, the number of sen-

0141-0296/03/$ - see front matter  2002 Elsevier Science Ltd. All rights reserved.
PII: S 0 1 4 1 - 0 2 9 6 ( 0 2 ) 0 0 1 1 8 - 9
58 J.-T. Kim et al. / Engineering Structures 25 (2003) 57–67

sors and the choice of sensor coordinates may have a For the MDOF structural system of NE elements and
crucial effect on the accuracy of the damage detection N nodes, the damage inflicted at predefined locations
procedure. may be predicted using the following sensitivity equ-
The objective of this paper is to present and evaluate ation [7]:
two methodologies to nondestructively locate and size

NE
damage in structures for which a few (1–3) natural fre- Fijaj ⫽ Zi (1)
quencies and/or mode shapes are available. First, a fre- j⫽1
quency-based damage detection (FBDD) method is out-
lined. A damage-localization algorithm to locate damage in which aj(⫺1ⱕajⱕ0) is the damage inflicted at the jth
from changes in natural frequencies and a damage-sizing location, i.e., the fractional reduction in jth element’s
algorithm to estimate crack-size from natural frequency stiffness parameter. The term Zi is the fractional change
perturbation are formulated. Next, a mode-shape-based in the ith eigenvalue due to damage, if changes in mass
damage detection (MBDD) method is outlined. A dam- are neglected, is given by:
age index algorithm to localize and estimate severity of Zi ⫽ dw2i / w2i (2)
damage from monitoring changes in modal strain energy
is formulated. Finally, the FBDD method and the MBDD where dw2i ⫽ wi∗2⫺w2i. The term Fij is the fraction of
method are evaluated for several damage scenarios by modal energy (i.e., sensitivity) for the ith mode that is
locating and sizing damage in numerically simulated concentrated in jth element and is given by:
prestressed concrete beams using a few natural fre- {fi}T[Cj]{fi}
quencies and mode shapes of the lower modes. Fij ⫽ (3)
{fi}T[C]{fi}
where {fi} is the ith mode shape vector, [C] is the sys-
2. Theory of nondestructive damage detection tem stiffness matrix, and [Cj] is the contribution of the
jth element to the system stiffness.
The NDD methodology presented here is designed to Once the quantity Zi is experimentally determined and
yield information on the location and the severity of the sensitivity Fij is numerically generated, Eq. (1) can
damage in a structure directly from measured changes be solved to locate and size damage in the system. How-
in the modal characteristics of the structure. The modal ever, the inverse solution is possible only if the number
characteristics of interest here are natural frequencies of damage parameters is close to the number of modes
and mode shapes. Once two sets of modal parameters are (i.e., NE⬇NM) [10]. In the case when NE ⬎ ⬎ NM, the
measured, for the as-built structure and its corresponding system may become ill-conditioned and alternate
damaged state, the NDD schemes described here are methods to estimate damage parameters must be sought.
used to predict the damage location and to estimate the In an effort to overcome this difficulty, an algorithm was
severity of the damage at that location. formulated using a frequency-change ratio and a sensi-
tivity ratio that are based on earlier works presented
2.1. The FBDD method by [4,14].
Consider a structural system with NE elements (j ⫽
2.1.1. Damage-localization algorithm 1,2,…,q,…,NE) and a measured set of NM vibration
With reference to Fig. 1, suppose we are given an modes (i ⫽ 1,…,m,n,…,NM). Eq. (1) is rewritten for any
pristine (i.e., undamaged) MDOF structural system with two modes m and n (m ⫽ n), respectively. Dividing Eq.
ith natural frequency, wi, and ith mode shape, fi. Next, (1) for mode m by the Eq. (1) for mode n, gives:
assume that at some later time the structure is damaged
冘F a
NE
(e.g., as shown in Fig. 1) in one or more locations of mj j
the structure. The resulting characteristic equation of the Zm j⫽1
⫽ (4)
Zn
冘F a
NE
damaged structure yields, respectively, frequencies and nj j
mode shapes w∗i and fi∗. Note that the asterisk denotes j⫽1

the damaged state. Fm1a1 ⫹ Fm2a2 ⫹ % ⫹ Fmqaq ⫹ % ⫹ FmNEaNE



Fn1a1 ⫹ Fn2a2 ⫹ % ⫹ Fnqaq ⫹ % ⫹ FnNEaNE

Assuming that the structure is damaged at a single


location q, such that aj ⫽ 0 when j ⫽ q but aj ⫽ 0
when j ⫽ q, Eq. (4) reduces to:
Zm / Zn ⫽ Fmq / Fnq (5)
In which Zm / Zn is the ratio of the fractional change
Fig. 1. Geometry of beam with a crack. in mth eigenvalue to the fractional change in nth eigen-
J.-T. Kim et al. / Engineering Structures 25 (2003) 57–67 59

value. Note also that Fmq / Fnq is the ratio of the sensi- can be computed by implementing results from linear
tivity for mth mode and qth element to the sensitivity of elastic fracture mechanics. Assuming plane strain con-
nth mode and qth element. Thus the damage inflicted at dition in the cracked beam, the energy loss rate of the
that location is defined by Eq. (5) equaling the L.H.S to ith modal strain energy is given by:
the R.H.S.
∂dWi (1⫺v2) 2
For all measured NM modes, Eq. (5) can be ⫽t KI (11)
extended into: ∂a E

冘 冘
NM NM where ∂dWi / ∂a is the energy loss rate of the ith modal
Zm / Zk ⫽ Fmq / Fkq (6) strain energy Wi with respect to crack depth a; t is the
k⫽1 k⫽1 beam thickness; n is Poisson’s ratio; and KI is the stress
Since Eq. (6) is true only if element q is damaged, intensity factor depending on crack depth a, applied
we introduce an error index into Eq. (6) as follows: flexural stress level s, and beam dimension (e.g., thick-
ness t, height H and length L).

冘 冘
NM NM
For the case of an edge-crack under bending (e.g., as
eij ⫽ Zm / Zk⫺Fmq / Fkq (7) shown in Fig. 1), the stress intensity factor is given by:
k⫽1 k⫽1

where eij represents localization error for the ith mode KI ⫽ F·s冑pa (12)
and the jth location, and eij ⫽ 0 indicates that the dam-
The term F is a geometrical factor depending on the
age is located at the jth location using the ith modal
dimensionless crack-length/beam-depth ratio a/H and
information. To account for all available modes we form
F ⫽ 1.12 for small cracks. Substituting Eq. (12) into Eq.
a single damage indicator (DI) for the jth member as:
(11) and further integrating Eq. (11) over the crack con-

冘 tour yields:
NM

DIj ⫽ [ e2ij]⫺1/2
冉 冊
(8)
i⫽1 pt(1⫺v2) 2 2 2
dWi ⫽ F skak (13)
where 0ⱕDIj ⬍ ⬁ and the damage is located at element 2E i

j if DIj approaches the local maximum point. in which, for the ith mode, ak ⫽ a(xk) represents the
crack size at location xk and sk ⫽ s(xk) represents the
2.1.2. Damage-sizing algorithm maximum flexural stress at location xk along the beam’s
Assuming no volume changes due to cracks or other longitudinal axis. For the Euler–Bernoulli beam, the
geometrical changes, Gudmunson [21] proposed a first stress level is given by:
order perturbation method that predicts the changes in
natural frequencies of a structure resulting from the dam- 1
s(xk) ⫽ EHf⬙i(xk) (14)
age. According to Gudmunson [21], for small cracks, the 2
fractional changes in modal strain energy can be related
to the fractional changes in frequency as follows: On dividing Eq. (13) by Eq. (10), the fractional
change in the ith modal strain energy is given by:
dWi dw2i
⫽ 2 (9) dWi pt(1⫺v2)H2 2
Wi wi ⫽ F Sika2k (15)
Wi 4 I
where Wi is the ith modal strain energy of the initial
structure, dWi is the loss in the ith modal strain energy in which Sik represents the sensitivity of the kth location
after damage, and dw2i / w2i is the fractional change in the in the ith modal strain energy and is given by:
ith eigenvalue due to the damage.

冕 冕
L
In the present study, we limit our discussion of the
crack-size model to Euler–Bernoulli beams. If the Euler– Sik ⫽ {f⬙i}2dx / {f⬙i}2dx (16)
k
Bernoulli beam theory is used, the ith modal strain 0
energy Wi is given by: On substituting Eqs. (10) and (15) into Eq. (9), The


L following relationship between the crack depth and the
1 fractional changes in the ith eigenvalue is obtained:
Wi ⫽ EI{f⬙i(x)}2dx (10)

冉冊
2
0 dw2i ak 2
⫽ hSik (17)
where E is Young’s modulus, I is the second moment w2i H i

of area, L is the beam span length, and fi(x) is the ith where for the beam section considered here,
mode shape function. Next, the change in the ith modal
strain energy due to the crack (e.g., as shown in Fig. 1) h ⫽ 0.25pt(1⫺v2)F2H4I⫺1 (18)
60 J.-T. Kim et al. / Engineering Structures 25 (2003) 57–67

(ak/H)i is the dimensionless crack size at the kth location of modal energy at the jth member and for the ith mode.
defined in the ith mode, and h is a constant which On differentiating Eq. (22), the quantity dFij is given by:

冋 册
depends upon beam dimensions, the crack type, and the
value of Poisson’s ratio. Eq. (17) can be solved to esti- Kij dKij dKi
dFij ⫽ ⫺ (25)
mate crack sizes if the quantities dw2i / w2i and Sik are Ki Kij Ki
experimentally determined or numerically generated.
Assuming Ki ⬎ ⬎ Kij (when NE ⬎ ⬎ l), Eq. (25)
can be further simplified:
2.2. MBDD method
dKij
dFij⬵ (26)
Consider a homogeneous, uniform cross-sectional, Ki
one-dimensional beam with NE elements (in the finite
Thus the quantity dFij in Eq. (26) can be measured
element sense) and N nodes. Assume that the input–out-
directly from changes in modal parameters.
put relationship of the beam is linear. Assuming a sol-
Assuming the structure is damaged in ND multiple
ution of the associated dynamic eigenvalue problem, the
locations, dKij⬇dKi / ND. The quotient dKij/Ki can be
ith modal stiffness, Ki, of the beam is given by:
approximated by the fractional change in the ith eigen-


L value due to damage by [7]:
Ki ⫽ k(x)[f⬙i(x)]2dx
0
(19)
gi ⫽
dKij 1 dw2i

Ki nd w2i 冉 冊 (27)

where fi(x) is the mode shape of ith modal vector and in which gi is a dimensionless factor representing the
k(x) is the bending stiffness of the beam (i.e., the product fractional change in the ith modal parameters. The term
of Young’s modulus and the second moment of area). dKij represents the variation of the modal stiffness and
The contribution of the jth element to the ith modal stiff- from Eqs. (20) and (24), it is given by:
ness, Kij, is given by:
dKij ⫽ K∗ij ⫺Kij ⫽ g∗ij kj∗⫺gijkj

(28)
Kij ⫽ kj [f⬙i(x)]2dx
j
(20)
冕 冕
where gij ⫽ [f⬙i(x)]2dx and g∗ij ⫽ [f⬙∗j (x)]2dx. Next, by
j j
where kj is the stiffness of the jth element and the inte- assuming that Young’s modulus and the second moment
gral is over the beam length. of area are constant over the entire beam (i.e., k(x) ⫽
The fraction of the modal energy for the ith mode that k(x̂) for 0ⱕxⱕL), and further on approximating the
is concentrated in jth member (i.e., sensitivity of the ith stress distribution in Eq. (19) by kj ⫽ k(x̂) (for j ⫽ 1,
modal stiffness with respect to jth element) is given NE), the right-hand side of Eq. (26) can be rewritten as:
by [22]:
dKij k∗j gij∗⫺kjgij
Fij ⫽ Kij / Ki (21) ⫽ (29)
Ki kjgi
Let the corresponding modal parameters in Eqs. (19)


L
and (21) associated with the damaged structure be
characterized by asterisks. Then for the damaged struc- where gi ⫽ [f⬙i(x)]2dx.
ture 0
On equating Eqs. (27) and (29), the relative change
F ⫽ K / K ⫽ Fij ⫹ dFij

ij

ij

i (22)
in jth element stiffness is given by:
where scalars K∗ij and K∗i are given by:


L


K∗ij ⫽ k∗j [f⬙i∗(x)]2dx (23)
gi ⫽ [f⬙i(x)]2dx
0
(30)

j
To account for several measurable and identifiable
and modes, the damage localization index for the jth
location, bj, in Eq. (30) is given by:


L

K∗i ⫽ k∗[f⬙i∗(x)]2dx
0
(24) bj ⫽
kj
kj∗
⫽ 冘 冘
i
gij∗ /
i
(gigi ⫹ gij) (31)

The term dFij represents the variation of the fraction where damage is indicated as the relative change in jth
J.-T. Kim et al. / Engineering Structures 25 (2003) 57–67 61

element stiffness when the L.H.S. of Eq. (31) is greater n=0.18, and r=2300 kg/m3; and (2) for steel tendon
than one. elements, the elastic modulus, Poisson’s ratio’s and mass
Once damage is located at the jth element, the severity density were taken to be E ⫽ 210GPa, n ⫽ 0.3, and
of damage is estimated at the same element directly from r ⫽ 7850 kg / m3, respectively.
Eqs. (30) and (31). The severity estimation index for jth The pre-damage and post-damage modal parameters
location, aj, is given by: of the FE model were generated numerically using the

冘 冘 冘
commercial software ANSYS. Here, six damage cases
kj∗⫺kj
aj ⫽ ⫽( (gij⫺gij∗) ⫹ gigi) / g∗ij (32) were investigated, as summarized in Table 1. As shown
kj i i i in Figs. 2 and 3, damage was simulated by a designed
where the damage severity is the fractional change in crack-depth for each damage case and was inflicted by
stiffness of the jth element. eliminating the stiffness of the appropriate elements in
the FE model. Note that the thickness of the crack was
defined as the length of an element (i.e., 1.25-cm). The
3. Numerical experiment first three damage cases CL1~CL3 (listed in Table 1)
were simulated by the cracks inflicted at 0.89m location
3.1. Description of test structure (e.g., x / L ⫽ 0.248) from the left edge which is in the
vicinity of the left quarter-span. The remaining three
A numerical experiment was performed to evaluate damage cases CM1~CM3 (listed in Table 1) were simu-
the proposed FBDD and MBDD methods. A simply sup- lated by the cracks inflicted at 1.79m location (e.g.,
ported PS concrete beam was selected (see [23] for x / L ⫽ 0.498) which is the vicinity of the mid-span.
detailed explanation on the target structure) and modal The extracted modal parameters of the test structure
responses of the structure were generated using finite included the pre-damage and post-damage frequencies
element models before and after damaging episodes. As and mode shapes of the first two bending modes. The
shown in Fig. 2, the test structure was the FE model of natural frequencies for the undamaged state and the six
the PS concrete beam of length L ⫽ 3.6 ⫺ m and the damage cases are listed in Table 1. The mode shape vec-
rectangular cross-section t × H ⫽ 0.1 ⫺ m × 0.125 ⫺ tors were read at 11 locations that are equally spaced
m. For modal analysis purposes the beam was divided along the top centerline of the beam’s longitudinal axis
into 28,512 block elements. A steel tendon was modeled (e.g., 36 cm between two adjacent locations). The
in the center of the beam by 288 elements and each undamaged mode shapes of the test structure are shown
element size was 1 cm × 1.25 cm × 1.25 cm. All other in Fig. 4. Also, the undamaged mode shapes were com-
elements modeled the concrete and each element size pared to the post-damage mode shapes as shown in Figs.
was 1.125 cm × 1.125 cm × 1.25 cm. The material 5 and 6. Fig. 5 shows the mode shapes of the modes 1
properties of the FE model were assigned as follows: (1) and 2 obtained before and after the cracks were inflicted
for the concrete elements, the elastic modulus, Poisson’s in the left quarter-span (i.e., CL1~CL3). Fig. 6 shows
ratio’s and mass density were taken to be E ⫽ 25GPa, the mode shapes of the modes 1 and 2 obtained before
and after the cracks were inflicted at the mid-span (i.e.,
CM1~CM3). Note from the figures that the amplitude
changes in the mode shapes alone are not sensitive
enough to monitor and localize damage at the inflicted
locations.

3.2. Damage detection by FBDD method

The Euler–Bernoulli beam model was selected as the


damage detection model (DDM), i.e., the mathematical
representation for damage detection practice. Modal
parameters needed for the FBDD process are pre-dam-
age and post-damage natural frequencies and pre-dam-
age mode shapes. The DDM of the structure consists of
a total of 288 beam elements of equal size, as schemat-
ized in Fig. 7 (note that the DDM elements 1~288). Each
DDM element is a potential damage location and has a
spacing of 1.25-cm or 0.347% (i.e., 1 / 288 × 100) of the
beam span. We justify the use of a 1.25-cm wide element
by interpolating measured modal vectors at the 289
Fig. 2. Schematic of test structure. nodal points of the damage detection model obtained by
62 J.-T. Kim et al. / Engineering Structures 25 (2003) 57–67

Table 1
Damage scenarios and natural frequencies of test beam

Damage case Inflicted damage Natural frequency (Hz)


Location (x/L) Crack-size (a/H) Severity (⌬EI/EI) Mode 1 Mode 2

Reference – – – 11.196 44.074


CL1 0.248 0.09 ⫺0.24 11.162 43.857
CL2 0.248 0.27 ⫺0.59 11.019 42.701
CL3 0.248 0.45 ⫺0.78 10.746 40.825
CM1 0.498 0.09 ⫺0.24 11.142 44.074
CM2 0.498 0.27 ⫺0.59 10.849 44.072
CM3 0.498 0.45 ⫺0.78 10.350 44.064

Fig. 3. Damage simulation to test structure.

Fig. 4. Pre-damage mode shapes of test structure.

Fig. 5. Comparison of mode shapes: undamaged vs damage cases


the use of spline functions and the element modal ampli-
CM1-CM3.
tude values from the mode shapes (i.e., Fig. 4) of the
FE model. Using the interpolated modal coordinates for
the beam, we generated functions f(x), where x is the
coordinate along the axis of the beam.
The modal sensitivity (i.e., the equivalent expression Since two natural frequencies are available, the sensi-
of Eq. (3)) of mode i and element j between two tivities are defined for 2 modes and 288 DDM elements.
locations (xj, xj+i) was computed by [22]: Fig. 8 shows the modal sensitivities of the test beam that
were computed along the beam’s longitudinal axis. It is

冕 冕
xj+1 l
observed that either the individual modal sensitivities or
dx
Fij ⫽ EI{f⬙i(x)}2 ;Ki ⫽ EI{f⬙i(x)}2dx (33) the combined sensitivities are indicative for most
Ki locations throughout the span except for the both ends
xj 0
of the simply supported beam. The maximum sensitivity

The curvatures f (x) of the mode shapes were gener-
i of the 1st mode is at the mid-span and that of the 2nd
ated at the 289 nodes of the damage detection model. mode at the quarter-span. Note that the single second
J.-T. Kim et al. / Engineering Structures 25 (2003) 57–67 63

Fig. 9. Error indices for damage cases CL1-CL3.

mode is not sensitive to damage locations near the


mid-span.
The fractional changes in frequencies (i.e., Eq. (2))
were computed by using the frequency results listed in
Table 1. By assuming that the flexural rigidity EI is con-
stant over the beam span, the sensitivity ratio (i.e., the
right-hand side of Eq. (17)) for an element q and for any
two modes m and n can be rewritten by:

冕 冕
l
2
{f⬙m} dx {f⬙ (x)}2dx
n
Fmq q

0


·l (34)


Fig. 6. Comparison of mode shapes: undamaged vs damage cases Fnq
CL1-CL3. 2
{f⬙n} dx
{f⬙m(x)}2dx
q 0

Next, localization errors were computed using Eq. (7)


for 2 modes and 288 locations (i.e., e1j and e2j, j ⫽
1,288) by implementing the sensitivity ratios and the
fractional changes in frequencies. Error indices of mode
1 for the damage cases CL1~CL3 are plotted in Fig. 9
and error indices of mode 1 for CM1~CM3 are plotted
Fig. 7. Schematic of damage detection model.
in Fig. 10. In the Figures note that each point where
error equals to zero indicates that a crack is located at
that location. Finally, we computed the single damage

Fig. 8. Modal sensitivities of test structure. Fig. 10. Error indices for damage cases CM1-CM3.
64 J.-T. Kim et al. / Engineering Structures 25 (2003) 57–67

simulated element. Also, in Fig. 12 the crack was simu-


lated at the mid-span and the location coincides with the
DDM element 144. The predicted peak in Eq. (8) was
close to the location of the simulated damage.
The accuracy of the damage localization presented
here was evaluated by measuring the so-called localiz-
ation error le ⫽ (⌬x / L) × 100, in which ⌬x is the metri-
cal difference between the real crack location and the
predicted location and L is the reference span. From a
comparison of the true and predicted locations, it was
observed that the minimum localization error is 0.3 per-
cent (damage case CM1) and the maximum localization
Fig. 11. Damage localization results of FBDD method: damage case
CL2.
error is 3.8 percent (damage case CL3) by excluding the
false-positive prediction. These results indicate that the
predicted locations fall within 1-cm~13.7-cm of the cor-
rect locations in the test beam (note that L ⫽ 360 ⫺
cm).
The size of damage size at each predicted location
was estimated by using the damage-sizing model given
by Eq. (17). Assuming a crack is located in element
between two locations (xj, xj+1), solving for the crack
size gives:

冉冊
ak
H i
⫽ 冪h·S ·l
dli
ik i
(35)

where (ak/H)i is the dimensionless crack size estimated


at location xk( ⫽ (xj ⫹ xj ⫹ 1) / 2)) by using the ith modal
Fig. 12. Damage localization results of FBDD method: damage case data. The modal sensitivity of mode i and location k was
CM2. computed using Eq. (33). The constant n was obtained
from Eq. (18) with H ⫽ 0.125 m, L ⫽ 3.6 m, Poisson’s
ratio of 0.18, and the geometrical factor F ⫽ 1.12. The
index given by Eq. (8) to decide potential crack fractional changes in the eigenvalues were computed
locations. from the pre-damage and post-damage frequencies listed
For example, damage indices of damage cases CL1 in Table 1. The damage-sizing results for the six damage
and CM1 are plotted in Figs. 11 and 12, respectively. In cases are summarized in Table 2.
the Figures, note that because of symmetry two predic- The accuracy of the damage sizing was evaluated by
tions are made thus one location is a false-positive pre- measuring the size error, which represents the difference
diction. The damage localization results for the six dam- between real and predicted crack-depths. It was observed
age cases are summarized in Table 2. Note that the that for damage cases CL1-CL3 the size errors ranged
prediction was made by either indicating DDM elements from 27.7 to 56.6% and for CM1-CM3 the size errors
or dimensionless locations. In Fig. 11 the crack was ranged from 2.2 to 22%. Note that a 1% error represents
simulated at the quarter-span and it is identical to the 0.1-mm difference in the estimation of 10-mm crack-
DDM element 72. The predicted peak was close to the depth.

Table 2
Damage prediction results of test beam using FBDD method

Damage case Inflicted damage Predicted damage


DDM Element Location (x/L) Crack-size (a/H) DDM Elements Location(s) (x/L) Crack-depth (a/H)

CL1 72 0.248 0.09 69,219 0.238, 0.759 0.141, N/A


CL2 72 0.248 0.27 66, 217 0.227, 0.752 0.358, N/A
CL3 72 0.248 0.45 61,213 0.210, 0.738 0.574, N/A
CM1 144 0.498 0.09 143, 144 0.495, 0.498 0.110, 0.110
CM2 144 0.498 0.27 142, 145 0.491, 0.502 0.276, 0.276
CM3 144 0.498 0.45 141 0.488 0.420, 0.420
J.-T. Kim et al. / Engineering Structures 25 (2003) 57–67 65

Fig. 13. Modal sensitivity of 1st mode: pre-damage vs damage cases Fig. 15. Modal sensitivity of 1st mode: pre-damage vs damage cases
CL1~CL3. CM1~CM3.

3.3. Damage detection by MBDD method

The Euler–Bernoulli beam model was selected as the


damage detection model for the MBDD method. As
described previously (i.e., as shown in Fig. 8), the DDM
of the structure consists of a total of 288 beam elements
of equal size. Modal parameters needed in the MBDD
process are pre-damage and post-damage mode shapes
and frequencies. For individual mode shapes, pseudo
readings at the 289 nodal points of the damage detection
model were obtained via spline interpolation functions.
Using the interpolated modal coordinates for each mode
shape, functions f(x) and f”(x) were generated, where Fig. 16. Modal sensitivity of 2nd mode: pre-damage vs damage cases
x is the coordinate along the axis of the beam. Then the CM1~CM3.
pre-damage and post-damage modal sensitivities were
computed, as shown in Figs. 13–16. Figs. 13 and 14
show, respectively, the 1st and 2nd modal sensitivities Next, the damage localization index (i.e., Eq. (31)) of
of the test beam before and after the damage cases CL1- element j was computed for the six damage cases. Then
CL3. Figs. 15 and 16 show, respectively, the 1st and 2nd the values of the indicator was normalized according to
modal sensitivities before and after the damage cases the rule:
CM1–CM3. In Figs. 13–16 note that changes in the
Zj ⫽ (bj⫺mβj) / sβj (36)
modal sensitivities are indicative to damage inflicted in
the mid-span (see Figs. 13 and 14) and to damage where mβj is the Mean of bj and sβj is the standard devi-
inflicted in the left quarter-span (see Fig. 15). ation of bj. The beam elements were next assigned to a
damage class via a statistical-pattern-recognition tech-
nique that utilizes hypothesis testing. The null hypothesis
(i.e., H0) was taken to be “the structure is undamaged at
the jth element” and the alternate hypothesis (i.e., H1)
was taken to be “the structure is damaged at the jth
element”. In assigning damage to a particular location,
the following decision rule were utilized: (1) choose H1
if Zjⱖ2 and (2) choose H0 otherwise. This test corre-
sponds to a confidence level of 0.98.
The damage localization results for the six damage
cases are listed in Table 3. For example, damage localiz-
ation indices of damage cases CL2 and CM2 are plotted
in Figs. 17 and 18, respectively. In the latter figures, the
Fig. 14. Modal sensitivity of 2nd mode: pre-damage vs damage cases damage indices are greater than two in a region near to
CL1~CL3. and including the peak locations. Note that in Fig. 17
the crack was simulated at the quarter-span and the
66 J.-T. Kim et al. / Engineering Structures 25 (2003) 57–67

Table 3
Damage prediction results of test beam using MBDD method

Damage case Inflicted damage Predicted damage


DDM Location Severity Range of DDM Most Probable DDM Element Severity (⌬EI/EI)
Element (x/L) (⌬EI/EI) Elements [Location] [Location]

CL1 72 0.248 ⫺0.24 55~81 [0.191–0.280] 67 [0.233] ⫺0.074


CL2 72 0.248 ⫺0.59 55~80 [0.191–0.277] 67 [0.233] ⫺0.364
CL3 72 0.248 ⫺0.78 55~80 [0.191–0.277] 67 [0.233] ⫺0.509
CM1 144 0.498 ⫺0.24 133~155 [0.48–0.52] 144 [0.498] ⫺0.132
CM2 144 0.498 ⫺0.59 133~155 [0.48–0.52] 144 [0.498] ⫺0.538
CM3 144 0.498 ⫺0.78 133~155 [0.48–0.52] 144 [0.498] ⫺0.788

The 1.54% error means that the predicted location devi-


ates 6.25-cm from the correct location in the 3.6-m
span length.
Once the damage was located, the severity estimation
index (i.e., Eq. (32)) was computed for the predicted
location. For each damage scenario, a damage-severity
value was estimated at the peak locations listed in Table
3. The accuracy of the damage sizing is evaluated by
measuring the size error that is the difference between
inflicted and predicted severities of damage. It is
observed that for damage cases CL1–CL3 the size errors
ranged from 34.7 to 69.0% and for CM1–CM3 the size
Fig. 17. Damage localization results of MBDD method: damage case
CL2. errors ranged from 1.0 to 45%.

4. Summary and conclusions

This paper presented a methodology to nondestruc-


tively locate and estimate the size of damage in struc-
tures for which two frequencies and/or mode shapes
were available. First, a frequency-based damage detec-
tion (FBDD) method was outlined. A damage-localiz-
ation algorithm that locates damage from changes in
natural frequencies and a damage-sizing algorithm that
estimates crack-size from natural frequency perturbation
were formulated. Next, a mode-shape-based damage
detection (MBDD) method was outlined. A damage-
index algorithm that locates and estimates severity of
damage from changes in modal strain energy was formu-
Fig. 18. Damage localization results of MBDD method: damage case lated. The FBDD method and the MBDD method were
CM2. evaluated for several damage scenarios by locating and
sizing damage in numerically simulated prestressed con-
crete beams for which only two sets of modal parameters
location is the DDM element 72. The predicted peak were available. For the verification test, natural fre-
element was close to the simulated element. Note also quencies and mode shapes of the first two bending
that in Fig. 18 the crack was simulated at the mid-span modes were generated from finite element models.
and the location is identical to the DDM element 144. By applying the FBDD approach to the test structure,
The predicted peak element was identical to the simu- it was observed that damage could be located with a
lated element. There is no damage in any other part of relatively small localization error. The predicted
the structure. The accuracy of the damage localization locations fell within 1-cm (i.e., the cracks near the mid-
is evaluated by measuring the localization error le. It is span) and within 13.7-cm (i.e., the cracks near the left
observed that the localization errors ranged from 0.0% quarter-span) of the correct locations in the 3.6-m beam
(CM1, CM2 and CM3) to 1.54% (CL1, CL2, and CL3). span. It was also observed that the size of crack could
J.-T. Kim et al. / Engineering Structures 25 (2003) 57–67 67

be estimated accurately for the crack located at the mid- ation in Solids, Int J. Analytical and Experimental Modal Analy-
span. The accuracy of the predicted crack size decreased sis 1990;5(2):67–79.
[8] Wu X, Ghaboussi J, Garrett JH. Use of Neural Networks in
for the damage at the quarter-span. By applying the Detection of Structural Damage. Computers and Structures
MBDD approach to the test structure, it was observed 1992;42(4):649–59.
that damage could be located accurately. The predicted [9] Kaouk M, Zimmerman DC. Structural Damage Assessment
locations were identical to the inflicted locations (i.e., Using a Generalized Minimum Rank Perturbation Theory. AIAA
the cracks near the mid-span) and were within 6.25 cm J 1994;32(4):836–42.
[10] Stubbs N, Kim JT. Damage Localization in Structures Without
from the correct locations (i.e., the cracks near the left Baseline Modal Parameters. AIAA Journal 1996;34(8):1649–54.
quarter-span) in the 3.6-m beam span. It was also [11] Liew KM, Wang Q. Application of Wavelets for Crack Identifi-
observed that the severity of the damage could be esti- cation in Structures. Journal of Engineering Mechanics, ASCE
mated accurately for the cracks located at the mid-span 1998;124(2):152–7.
and the prediction accuracy decreased for the cracks at [12] Lee YY, Liew KM. Detection of Damage Locations in a Beam
using the Wavelet Analysis. International Journal of Structural
the quarter-span. Stability and Dynamics 2001;1(3):455–65.
[13] Ostachowicz WM, Krawczuk M. Vibration Analysis of a Cracked
Beam. Computers & Structures 1990;36(2):245–50.
Acknowledgements [14] Stubbs N, Kim JT, Topole K. The Effect of Model Uncertainty
on the Accuracy of Global Nondestructive Damage Detection in
Structures. In: Spanos PD, Brebbia CA, editors. Computational
This study was supported by Korea Science and Stochastic Mechanics. London: Elsevier Applied Science; 1991.
Engineering Foundation (KOSEF) through Smart Infra- p. 125–36.
Structure Technology Center in the program year of [15] Sundermeyer, J.N., Weaver, R.L., 1993, On Crack Identification
2002. and Characterization in a Beam by Nonlinear Vibration Analysis,
TAM Report No. 74, UILU-ENG-93-604, Univ. of Illinois.
[16] Kim J.T., 1993, Assessment of Relative Impact of Model Uncer-
tainty on the Accuracy of Nondestructive Damage Detection in
References Structures, Ph.D. Dissertation, Texas A&M University, USA
[17] Chen J, Garba JA. On-Orbit Damage Assessment for Large Space
[1] Rytter, A., 1993, Vibration Based Inspection of Civil Engineer- Structures. AIAA Journal 1988;26(9):1119–26.
ing, Ph.D. Dissertation, University of Aalborg, Denmark. [18] Pandey AK, Biswas M. Damage Detection in Structures using
[2] Vandiver, J.K., 1975, Detection of Structural Failures on Fixed Changes in Flexibility. Journal of Sound and Vibration
Platforms by Measurement of Dynamic Responses, Proc. of 7th 1994;169:3–17.
Annual Offshore Technology Conference, Houston, Texas, [19] Farrar CR, Jauregui DA. Comparative Study of Damage Identifi-
Paper 2267. cation Algorithm Applied to a Bridge: I. Experiment, Smart
[3] Crohas H, Lepert P. Damage Detection Monitoring Method for Mater. Struct 1998;7:704–19.
Offshore Platforms Is Field Tested. Oil and Gas J. [20] Kim, J.T., Jung, S.H., Lee, Y.K., Yun, J.W., 2000, Damage
1982;80(8):1982. Identification in Bridges using Vibration-Based System Identifi-
[4] Cawley P, Adams RD. The Location of Defects in Structures cation Scheme, Proc. of 18th Int. Modal Analysis Conference,
from Measurements of Natural Frequencies. J. Strain Analysis San Antonio, Texas, pp. 1327-1333.
1979;14(2):49–57. [21] Gudmunson P. Eigenfrequency Changes of Structures due to
[5] Pandey AK, Biswas M, Samman MM. Damage Detection from Cracks, Notches or Other Geometric Changes. J. Mech. Phys.
Changes in Curvature Mode Shapes. Journal of Sound and Solids 1982;30(5):339–53.
Vibration 1991;145(2):321–32. [22] Kim JT, Stubbs N. Model Uncertainty and Damage Detection
[6] Chance, J., Tomlinson, G.R., Worden, K., 1994, A Simplified Accuracy in Plate-Girder Bridges. Journal of Structural Engineer-
Approach to the Numerical and Experimental Modeling of the ing, ASCE 1995;121(10):1409–17.
Dynamics of a Cracked Beam”, Proc. of the 12th Int. Modal [23] Saiidi M, Douglas B, Feng S. Prestress Force Effect on Vibration
Analysis Conference, Honolulu, Hawaii, Vol. 1, 778-785. Frequency of Concrete Bridges. Journal of Structural Engineer-
[7] Stubbs N, Osegueda R. Global Nondestructive Damage Evalu- ing, ASCE 1994;120(7):2233–41.

You might also like