Third International Conference on

Advances in Control and Optimization of Dynamical Systems

March 13-15, 2014. Kanpur, India

Damage Characterization in Frame

Structures using Output-only Modal and
Feature-based Techniques
Harikrishnan Panikkaveettil ∗ , Koushik Roy ∗∗ ,
Samit Ray-Chaudhuri ∗∗∗
∗
Former Masters’ Student, Department of Civil Engineering, IIT
Kanpur, India- 208016, email:[email protected]
∗∗
Doctoral Student, Department of Civil Engineering, IIT Kanpur,
India- 208016, email:[email protected]
∗∗∗
Assistant Professor, Department of Civil Engineering, IIT Kanpur,
India- 208016, email: [email protected]

Abstract: A combined approach involving a modal-based as well as feature-based technique

has been proposed for damage characterization in frame structures. In modal-based approach,
the modal parameters of a structure are estimated from ambient responses using an output-
only system identification technique and subsequently the fundamental mode shape and
its derivatives are used for damage characterization purpose. In feature-based approach,
autoregressive models with exogenous input (ARX model) are built up using healthy and
damaged structural responses and the first ARX model coefficients are used for damage
characterization. Simulation studies are carried out on a two-dimensional three-bay twelve-
story steel moment-resisting frame to illustrate the applicability of the techniques in identifying
damages in beams and columns. Both modal and feature-based approaches are capable of
identifying the nature and extent of damages. However, an improvement of the algorithm is
desirable for damage quantification and prognosis.

Keywords: health monitoring, mode shape curvature, ambient vibration, output-only method,
autoregressive model.

1. INTRODUCTION inspection can then be carried out to ascertain the visible

verification of damage extent.
After a strong earthquake, many structures may undergo Structural damage detection of civil structures is a
damage with severity varying from minor damage to com- highly challenging problem. This is because of the var-
plete collapse. It is customary to assess the safety of ap- ied and complex nature of the challenges associated with
parently damaged structures by a well trained group of ex- this kind of structures. These challenges include difficul-
perts before the authorities allow these structures to be re- ties in exciting the structures because of their massive
occupied or opened for operation. This safety assessment is sizes, necessity for tests under natural and operational
usually performed through visual inspection, which is often conditions, unknown excitations, and uncertainties asso-
difficult, experience-based and time consuming. For many ciated with structural parameters. The research empha-
civil engineering structures (such as an office building or sis on vibration-based damage detection can widely be
a bridge), the structural components are often difficult to observed by a large number of articles [Doebling et al.,
access for the purpose of visual inspection. For example, 1998, Kerschen et al., 2006, Yan et al., 2007] published in
in a typical office building, the structural components last couple of decades. Rytter [1993] divided the damage
are often inaccessible due to non-structural components detection methodology into four levels: detection, localiza-
such as false ceiling and architectural components. It is tion, quantification and prognosis; where Sohn and Farrar
very hard to detect manually, if any damage takes place [2001] introduced another intermediate stage as damage
somewhere in any structural members. Besides, the current characterization as shown in Fig. 1. It can be seen that,
practice of structural damage detection is time consuming, although many methodologies have been developed so far
expensive, laborious and experience-based. This is where for this purpose, most of them often fail to perform satis-
the vibration-based damage detection gets its additional factorily in terms of damage characterization for real civil
significance. Recent studies [Lourens et al., 2012] show that structures. Frame buildings form the most common class
a limited number of pre-installed sensors at strategic loca- of civil structures and not much of research is seen towards
tions would suffice this requirement of damage detection, detecting damage in such structures. Frame buildings are
localization and characterization. If damage is found, local generally designed following the weak-beam strong-column
? The work is funded by Ministry of Human Resources Development philosophy to ensure column damage prior to beam dam-
(MHRD), India. age.

978-3-902823-60-1 © 2014 IFAC 973 10.3182/20140313-3-IN-3024.00243

2014 ACODS
March 13-15, 2014. Kanpur, India

I t
Intermediate
di t
Level 1 Level 2 Level 3 Level 4
Level
Detection Localization Quantification Prognosis
Characterization

To predict
on
To check the To find the Extent or
Type of remaining
presence of location of Severity of
damage life and
d
damage d
damage d
damage
future
performance

Fig. 2. Difference in the influence of beam and column

damage
Fig. 1. Stages for structural damage detection
In this study, one modal-based and one feature-based
Since last couple of decades, a good number of research output-only method is used for detecting and character-
has been carried out on mode shape-based method in izing damage. Mode shape curvature and ARX model
damage detection. Pandey et al. [1991] proposed the con- coefficient are utilized as signature patterns to distinguish
cept of mode shape curvatures as indicators of damage among different damage cases. The proposed method cov-
in beams based on simulation studies. Abdo and Hori ers the first three levels of damage detection [Rytter, 1993].
[2002] used measured rotational mode shapes to localize A description has been provided to explain how beam and
damage in numerically simulated plates and beams. Maia column damage affect the structure in terms of extent of
et al. [2003] extended the concepts of mode shape based damage. Brief theoretical background has been discussed
methods for all frequencies by considering the vibrating to show the correlations of mode shape curvature and ARX
pattern for all frequencies and using them for damage model coefficient with the extent of damage. The efficiency
detection. Chakraverty [2005] identified the structural pa- of the method is studied through simulation studies on
rameters using limited modal information and limited a two-dimensional twelve-story three-bay steel moment-
sensor data in shear buildings by extending the work of resisting frame (SMRF).
Yuan et al. [1998]. Whalen [2008] studied the effect of
mass, stiffness and damage radius scale on mode shape 2. DAMAGE CHARACTERIZATION
derivatives of beam-like structures and showed that the
damage alters mode shape globally while the influence on A damage in a column or a beam may have different
mode shape derivatives are localized. Koo et al. [2010] implication towards overall vulnerability of a building.
suggested the use of damage induced inter-story drift for Therefore, in order to guarantee post-disaster serviceabil-
damage detection in shear buildings and demonstrated ity, characterization of damages between beam and col-
their use through numerical simulation and experimental umn is an important task. A parameter that generates
study. Dilena et al. [2011] applied modal curvature to similar signature pattern to all types of damages can not
detect damage in a three-span reinforced-concrete bridge. be a desired one to distinguish between a beam and a
Dos Santos et al. [2011] performed a numerical study on column damage. Therefore, it is important to understand
a beam to demonstrate the use of curvature difference, the physical phenomena of different types of damage and
damage index, modified curvature difference and modified its correlations with damage signatures.
damage index for damage localization. Fig. 2 describes the effect of different kind of damages
Feature-based methods attract the attention of the re- on structural stiffness. This stiffness alteration results in
searchers by virtue of their merits. Lu and Gao [2005] change of mode shape curvature [Roy and Ray-Chaudhuri,
derived the expression of ARX model parameters to cor- 2013] and ARX model coefficients Lu and Gao [2005].
relate with different structural parameters and also illus- When a beam is damaged, the influence is observed mostly
trated the significance of ARX model coefficients as well as on columns adjacent to that floor affecting that damaged
residual error in structural damage detection. Marec et al. degree-of-freedom (DoF) and also the consecutive DoFs;
[2008] investigated and characterize delamination in a whereas in case of a column damage (See Fig. 2), maximum
composite plate applying time-scale methods and wavelet influence takes place in the same story influencing only two
transform. De Lautour and Omenzetter [2010] extracted consecutive DoFs. Therefore, the effect of column damage
damage information from autoregressive model coefficients is more localized than that of beam damage. In other
to form damage sensitive features which was taken as words, the extent of beam damage on structural stiffness is
input in artificial neural network for damage classification more than the extent of column damage. Hence, formation
in a three-storey experimental building model. Discrete of peaks and their extents in mode shape curvature and
wavelet analysis was performed by Katunin [2011] to typify ARX model coefficients reflect the type of damage.
three distinguish damage scenarios in a layered composite
plate and this proposed methodology was also verified 3. THEORETICAL BACKGROUND
with experiments. Wu et al. [2012] employed instantaneous
frequency normalization and Hilbert-Huang transform in 3.1 Effect of damage extent on mode shape
gear fault diagnosis to distinguish among different types
of damages like worn tooth, broken tooth, gear unbalance Reinoso and Miranda [2005] conducted a detailed study
etc. An and Sohn [2012] proposed an improved and robust on six high-rise buildings to investigate the contribution
approach based on integrated impedance and guided wave of flexural as well as shear deformation on the overall
to identify different types of damage in a complex structure structural responses. Their studies showed that, even for a
by minimizing any false detection of damage. high rise SMRF, the shear contribution on the structural

974
2014 ACODS
March 13-15, 2014. Kanpur, India

the shake of numerical analysis. Fig. 3(a) shows the mode

shape for undamaged case and for two different lengths of
damage extents around y = 0.5, where total length of the
building is normalized to unity. The curve with solid line
corresponds to 10% and the dotted line corresponds to 15%
damaged portion of the total length. From this figure, it
can also be observed that at the damaged spot the mode
shapes does not get differed from each other, even they
do not show any visible distinction with the undamaged
mode shape. Fig. 3(b) shows the difference in mode shapes
(δφ(y)) of the damaged one from the undamaged mode
shape for both 10% and 15% damage extent cases. Fig. 3(c)
and Fig. 3(d) present the difference in mode shape slopes
(δφ0 (y)) and their curvatures (δφ00 (y)) profile along the
length of the beam. The difference in slope attains a max-
imum value at damaged point as shown in Fig. 3(c) and as
a result its second derivative changes its sign at the damage
location (see Fig. 3(d)). This results are in agreement with
the proposal made by Roy and Ray-Chaudhuri [2013].
Besides, a clear signature pattern is observed with the
extent of damage. It can be observed that in case of 15% of
damage extent, the change in curvature spread over more
length around the damage location in comparison to that
for 10% damage extent. This signature pattern is later on
used for damage characterization purpose because these
patterns are sensitive with the damage extents.

3.2 Autoregressive Model coefficients

The equation of motion of a n DoF system as shown in

Fig. 4 can be expressed as,

Fig. 3. (a) Mode shape, (b) difference in mode shape, (c)

difference in mode shape slope and (d) difference in [M0 ]¨{y(t)} + [K0 ]{y(t)} = {r0 }u(t) (2)
mode shape curvature and (e) the shear beam where {y(t)} is the structural response, [M0 ] and [K0 ] are
the mass and stiffness matrices respectively, {r0 } is the
response cannot be neglected. Therefore for a low rise influence matrix. This equation can be modified as,
structure, the building frame can be approximated as
shear building for mathematical modelling neglecting the
{ÿ(t)} + [K]{y(t)} = {r}u(t) (3)
rotational component of deformation [Chopra, 2007]. Such
kind of building can be idealized as a continuous shear [K] = [M0 ]−1 [K0 ] = structural normalized stiffness which
beam having infinite DoFs. To establish the effects on the can be termed as, [K] = [Φ]T [Λ][Φ], {r} = [M0 ]−1 {r0 } =
mode shape due to change in stiffness coefficients, consider normalized influence matrix, [Φ] and [Λ] are the mass nor-
free vibration of the prismatic shear beam (See Fig. 3(e)). malized mode shape and eigen value matrices respectively.
The normalized mode shape of the shear beam can be Autoregressive model with exogenous input (ARX model)
expressed as, are used in this study. The structural response in terms of
ARX model takes the following form Lu and Gao [2005],
πy 2 3
φ(y) = sin (1)
X X
2L ŷ(t) = [Ai ]y(t − i) + [Bj ]u(t − j) (4)
i=1 j=1
where L is the length of the beam. Now damage was
introduced by reducing the shear rigidity at midspan (y = where ŷ(t) = the prediction signal, [A1 ] = 2 cos (K 1/2 ∆t),
0.5L). Therefore, in order to investigate the influence of [A2 ] = −I, [B1 ] = [r], [B2 ] = −[Φ][I + cos (K 1/2 )][Φ]T [r]
different types of damage on mode shape, the shear beam and [B3 ] = [Φ] cos ([Λ]1/2 ∆t)[Φ]T [r] with ∆t and I being
has been damaged with various damage extents because the step size of the time series and identity matrix respec-
damages in a beam or a column alters the extent of damage tively. Evidently, the ARX model coefficients are functions
(i.e. affected DoFs or damage portion) differently. A 10% of structural parameters. Therefore, any sorts of deterio-
(damaged portion spanning from y = 0.45L to y = 0.55L) ration in structural properties will show up alteration of
and a 15% (damaged portion spanning from y = 0.425L ARX model properties.
to y = 0.575L) damage extents were considered.
Now considering first ARX model coefficients of an
The main purpose of the illustration is to show how mode undamaged and damaged structures, respectively, as A1u
shape curvature behaves with damage extent. Now, the and A1d , the difference in first ARX model coefficients can
shear beam has been discretized into 100 finite elements for be calculated as follows:

975
2014 ACODS
March 13-15, 2014. Kanpur, India

Fig. 5. Eigen system realization algorithm

Fig. 4. n-DoF spring mass system

NExT: If the response is stationary, it can be shown
2 that the cross-correlation function between the measured
[∆A1 ] = − ([Ku ] − [Kd ])∆t2
2! (5) responses can be used as the free response of the same
2 system [James et al., 1993, He and Fu, 2001, Shen et al.,
+ ([Ku ]2 − [Kd ]2 )∆t4 + − − −∞
4! 2003]. For a multi-degree-of-freedom system (MDoF), the
dynamic equation of motion can be expressed as,
Now in the expression of [∆A1 ] in Eq. 5, the higher order
term of ∆t2 in the infinite series can be neglected for very
small sampling period of ∆t and hence can be termed as, [M ]{ÿ(t)} + [C]{ẏ(t)} + [K]{y(t)} = {F (t)} (8)
where {ÿ(t)}, {ẏ(t)} and {y(t)} are acceleration, velocity
∆A1 = (Kd − Ku )∆t2 (6) and displacement vectors respectively; [M ],[C] and [K]
are the mass, damping and stiffness matrices respectively;
Now, let us consider that the damage took place at the {F (t)} is the excitation vector assumed to be taken as
p th story level of the building, which can be simulated as white signal, whose cross-correlation vanishes with a ref-
a stiffness degradation of that floor by an amount of δkp erence acceleration signal {Ryyi (τ )}. While taking cross-
from its original stiffness kp . The difference in ARX model correlation function of both side of Eq. 8, this results in
coefficients can be expressed in matrix form: the following differential equation,

··· ··· ···

 
0 0 0 0 [M ]{R̈yyi (τ )} + [C]{Ṙyyi (τ )} + [K]{Ryyi (τ )} = {0} (9)
0 0 0 0 ··· ··· 0
0 ··· ··· ··· 0 ··· 0
 
The above second order homogeneous differential equa-
[∆A1 ] =  0 0 · · · δkp ∆t2 −δkp ∆t2 0 0 (7)
 
tion of motion is comparable with the free vibration re-
0 ··· 0 −δkp ∆t2 δkp ∆t2 ··· 0 sponse of a MDoF system, where the cross-correlation
 
0 ··· ··· 0 ··· ··· 0 function {Ryyi (τ )} is analogically comparable with the free
0 ··· ··· 0 ··· 0 0 n×n vibration response of the structure with identical system
properties as the original structure.
Absolute value of [∆A1 ] will be a null matrix with
dimension n × n with four non-zero terms at element ERA: Researchers have reported successful application of
locations (p − 1, p − 1), (p − 1, p), (p, p − 1) and (p, p), ERA for problems from various fields like bridges [Heo
depicting the damage location at the pth level. Now in et al., 2004, Dionysius and Fujino, 2008, Quek et al., 1999],
case of damage at (p + 1)th level, the absolute value of buildings [Caicedo et al., 2004, Hedge and Sinha, 2008,
the difference between the first ARX model coefficients of Moaveni et al., 2011] etc. The algorithm (Fig. 5) takes
damaged and undamaged structure will bear again four the Markov parameters (impulse response in this context)
non-zero terms, but at element locations (p, p), (p, p + 1), as input. In this approach, a generalized Hankel matrix
(p + 1, p) and (p + 1, p + 1). As stated earlier, the extent [Juang and Pappa, 1985, He and Fu, 2001, Juang and
of damage in case of beam damage is more than that of Phan, 2001, Heo et al., 2004, Hedge and Sinha, 2008]
column damage. In other words, if beam damage took is constructed using the free vibration response of the
place, the affected DoFs are more than that of the column structure. Singular value decomposition of the Hankel
damage. Therefore, stiffness coefficients of more DoFs will matrix, [H(0)] = [U ][S][V ]T is used to determine the
be influenced resulting in more non-zero elements in the system matrix [A] and the output matrix [C] of the state
1 1
matrix [∆A1 ]. space model; [A] = [S]− 2 [U ]T [H(1)][V ][S]− 2 and [C] =
1
[U ][S] 2 . Finally, the modal parameters can be estimated
3.3 Output-only System Identification Method from the following expressions.

The system identification method used in this study is z = Ψ−1 AΨ & sn = ln z/dt (10a)
natural excitation technique coupled with eigen system |sn |
realization algorithm (NExT-ERA). Natural excitation fn = (10b)
2π
technique (NExT) is a very popular choice for finding the <sn
free vibration response from ambient vibration response ξ=− & Φ = CΨ (10c)
and eigen system realization algorithm (ERA) is used fn
for modal parameter estimation. These two techniques
together works as efficient output-only algorithms for where the eigen values and vectors of [A] be ω and Ψ
modal parameter estimation [Heo et al., 2004, Caicedo respectively. Also, fn , Φ and ξ are respectively natural
et al., 2004], where it is assumed that the structure is frequency, mode shape vector and damping ratio of the
linear and is being excited by a Gaussian white-noise. system and ∆t is the sampling period. However, it may be
Brief description of their algorithms are given here for the noted that the method requires free vibration response,
completeness of the paper. which can be obtained from NExT.

976
2014 ACODS
March 13-15, 2014. Kanpur, India

Fig. 6. 12 story building used for numerical study: (a) name

convention and (b) section details
4. NUMERICAL STUDIES

In this study, a three-bay twelve-story steel moment- Fig. 7. Various modal parameters: (a) difference in mode
resisting frame (SMRF) is considered. The frame was shape, (b) difference in slope of mode shape, (c)
previously designed by Santa-Ana and Miranda [1997]. change in curvature of mode shape
Fig. 6(a) shows the schematic diagram of the two- difference in slope (δφ0 (y)) is shown in Fig 7(b). From
dimensional model and the sections used in modelling the Fig 7(b), it can be observed that the plot corresponds
structure are shown in Fig. 6(b). The numerical model to beam damage case shows a peak at the location of
is developed in OpenSees version 2.4.1 [ope] using elastic damage. However, a similar peak with a scale factor is
beam-column elements. For the purpose of modelling, the noticed in case of column damage as well. Now, the differ-
structure has been assumed to be fixed at its base and a ence in mode shape curvature along the building height is
2% Rayleigh damping is considered for the first two modes. shown in Fig. 7(c), where very sharp and dominant peaks
A uniform seismic mass distribution along the height is are observed for both beam and column damage cases.
considered based on load calculation and masses of 64000 From Fig 7(c), one can observe that beam damage can be
kg per floor are lumped at nodal points at each floor level characterized by peaks in change of curvature plot at loca-
in such a way that the interior nodes are having twice tions corresponding to the floor levels immediately below
the mass of that of the exterior to mimic a realistic mass and above the floor containing damaged beam, whereas
distribution. An eigen value analysis is performed and the column damage can be characterized by peaks in change
first three frequencies are calculated as 0.65, 1.89 and 3.25 of curvature plot at locations corresponding to the floor
Hz. This is in agreement with those obtained by Ray- levels connected by the column. Therefore, any presence
Chaudhuri and Villaverde [2008]. of damage is characterized by the presence of dominant
To simulate two different damage scenarios (a column peaks in the vicinity of damage in the opposite directions
damage case and a beam damage case), a stiffness degra- (one is positive and the other is negative) without any
dation of 20% is introduced in all the columns of level 5-6 other dominant peak in between them. Hence, the effect
for column damage case and the same amount reduction of beam damage is more widespread in nature than that
in stiffness at all the beams at level 6 is introduced for of column damage and the location of these peaks can be
beam damage case, as shown in Fig. 6(a). The structure is used for identifying the nature of damage.
subjected to a base excitation using a band-limited white- Studies show that structural damage [Curadelli et al.,
noise to simulate ambient excitation. A time history analy- 2008, Montalvao et al., 2009], soil structure-interaction
sis is performed and horizontal floor acceleration responses [Novak and Hifnawy, 1983] and other non-structural
are recorded. changes [Miller, 2006] may raise the system damping.
Fig. 8(a) and Fig. 8(b) show the difference in mode shape
4.1 Damage Characterization using Modal Approach curvatures for column and beam damage cases respectively
with varying Rayleigh damping of 0%, 1%, 2% and 5% for
Modal parameters are evaluated with NExT-ERA using the first two modes. In both damage cases, the trend of
these ambient vibration data for damaged and undamaged variation of mode shape curvature is similar to that with
states. Estimated mode shapes are differentiated using Fig. 7(c).
central difference method to obtain mode shape curva-
tures. The results are presented as beam and column dam- Fig. 9 shows the mode shape curvature plot for various
age cases in order to characterize the types of damage in location of damage. Now, Figs. 9(a)-(c) show respectively
the structural model. The difference between the estimated the mode shape curvature corresponding to column dam-
damaged (column and beam damages) and undamaged ages at the Level 0-1, Level 2-3, Level 8-9 and to beam
mode shapes, i.e. δφ(y) are shown in Fig. 7(a) and their damages at the Level 1, Level 3 and Level 9. Since there

977
2014 ACODS
March 13-15, 2014. Kanpur, India

(a)

Fig. 8. Influence of damping on difference in mode shape

curvature: (a) storey column damage at the level 5-6
and (b) floor beam damage at the level 6

(b)

Fig. 10. Difference in ARX model coefficients for (a) storey

column damage at the level 5-6 and (b) floor beam
damage at the level 6
Fig. 9. Effect of location for damage at (a) level 0-1, (b) and beam damage case. Fig. 10(a) shows a 3D bar plot of
level 2-3, (c) level 8-9 for storey column damage and the difference of ARX model coefficients for the column
(a) level 1, (b) level 3, (c) level 9 for floor beam damage at the level 5-6. One can observe from this figure
damage that due to damage at the level 5-6, the 5th and the 6th
is no floor below floor Level 1, a slightly different trend floor levels are most affected, hence showing differences of
is observed in Fig. 9(a). It can be seen that the nature of the undamaged state from that of the damaged one in the
variation of change of curvature for all other damage loca- stiffness coefficients around the location of damage. Now,
tions follows the trend similar to Fig. 7(c), corresponding Fig. 10(b) shows another 3D bar plot of thethdifference of
to their respective locations of damage. ARX model coefficients for damage at the 6 floor beam
level. The figure depicts that due to the 6th floor beam
4.2 Damage Characterization using Feature-based Approach damage, the structural response adjacent to the damage
floor level get changed resulting in the difference in the
ARX model coefficients at the 5th , 6th and 7th storey level.
Since in case of ambient vibration case the input infor- The figure shows larger difference at the 5th and the 7th
mation is unknown, the output response of any one floor floor level. From Fig. 10, one can distinguish whether this
response can be considered as input Lu and Gao [2005]. In is a beam or column damage by looking at the extent of
this study, the DoF corresponding to Level 1 is assumed damage. In case of storey column damage, the adjacent two
as input and the responses of the remaining eleven DoFs floor responses alter more; whereas in case of floor beam
are considered as output to construct the autoregressive damage, the adjacent floors are more responsive than the
model with exogenous input (ARX model) model for dif- damaged floor.
ferent states and damage cases. ARX model coefficients are
determined from the model corresponding to undamaged Now, Figs. 11(a)-(c) shows the bar plot corresponding
state and damage state (beam and column damage). The to storey column damages at the level 1, level 3 and level
difference between first ARX model coefficients of the un- 9 respectively, whereas Figs. 11(d)-(f) are respectively for
damaged and damage states is estimated. The normalized floor beam damages at the level 1, level 3 and level 9.
absolute values of the difference are plotted against DoFs Fig. 11 depicts that regardless of damage location ARX
in the form of a three dimensional bar chart for column coefficient difference is capable of distinguishing types of

978
2014 ACODS
March 13-15, 2014. Kanpur, India

ture of damages can successfully be identified. However,

these methods need some improvements, since they are
not independent of the structure. Also, some new features
sensitive to severity of damage can be proposed for damage
quantification purpose. As the proposed approaches are
purely output-only demanding no input information, it
can be applied for real-life damage detection problems
involving frame buildings in identifying different kind of
damages.

REFERENCES
OpenSees- Open System for Earthquake Engineering
Simulation, Pacific Earthquake Engineering
Research Center (PEER), Richmond, CA, USA
(http://opensees.berkeley.edu/).
Fig. 11. Effect of location for column damage at (a) level 0- M. M. Abdel Wahab and G. D. Roeck. Damage detection
1, (b) level 2-3 and (c) level 8-9 and for beam damage in bridges using modal curvatures - Application to a real
at (d) level 1, (e) level 3 and (h) level 9 damage scenario. Journal of Sound and Vibration, 226
(2):217–235, 1999.
damages. Hence, damage characterization between beam M. A. B. Abdo and M. Hori. A numerical study of struc-
and column damage can be executed from the 3D bar plots tural damage detection using changes in the rotation of
of the difference in first ARX model coefficients of dam- mode shapes. Journal of Sound and Vibration, 251(2):
aged and undamaged state of the structure. However, Roy 227–239, 2002.
and Ray-Chaudhuri [2012] showed that the performance Y. K. An and H. Sohn. Integrated impedance and guided
of ARX model-based features decreases with increment of wave based damage detection. Mechanical Systems and
system damping and so the effect of damping on the ARX Signal Processing, 28:50–62, 2012.
model coefficients is not presented in this study. J. M. Caicedo, S. J. Dyke, and E. A. Johnson. Natural
excitation technique and eigensystem realization algo-
5. CONCLUSION rithm for phase I of the IASC-ASCE benchmark prob-
lem: Simulated data. Journal of Engineering Mechanics,
This study involved in proposing a combined (modal ASCE, 130(1):49–60, 2004.
as well as a feature-based) approach to distinguish be- S. Chakraverty. Identification of structural parameters of
tween different nature of damages in frame structures multistorey shear buildings from modal data. Earth-
using output-only measurements. In the modal approach, quake Engineering and Structural Dynamics, 34(6):543–
mode shapes are estimated from recorded ambient re- 554, 2005.
sponses using Natural Excitation Technique coupled with A. K. Chopra. Dynamics of Structures. Prentice-Hall of
Eigen System Realization Algorithm (NExT-ERA) and India Private Limited, 2nd edition, 2007.
the curvature of the fundamental mode shape is calcu- R. O. Curadelli, J. D. Riera, D. Ambrosini, and M. G.
lated with central difference approximations. Since various Amani. Damage detection by means of structural
damage scenarios alter these curvature values differently, damping identification. Engineering Structures, 30(12):
the differences in these curvature patterns are used for 3497–3504, 2008.
damage characterization. It is observed that the extent of O. R. De Lautour and P. Omenzetter. Damage classifi-
curvature change around the damage location is more in cation and estimation in experimental structures using
case of beam damage than that of column damage because time series analysis and pattern recognition. Mechanical
the beam damage comparatively affects more DoFs in a Systems and Signal Processing, 24(5):1556–1569, 2010.
structure. Difference in fundamental mode shape curva- M. Dilena, A. Morassi, and M. Perin. Dynamic identifica-
ture can distinguish damages between beam and column tion of a reinforced concrete damaged bridge. Mechan-
for wide range of damping values and irrespective of the ical Systems and Signal Processing, 25(8):2990–3009,
location of damages. Damage can also be characterized 2011.
between beam and column from the difference in first ARX M. S. Dionysius and Y. Fujino. System identification
model coefficients. The same structural responses were of suspension bridge from ambient vibration response.
used for autoregressive time series modelling to determine Engineering Structures, 30(2):462–477, 2008.
the model coefficients. In case of storey column damage S. W. Doebling, C. R. Farrar, M. B. Prime, and Shevitz D.
the adjacent floor responses are affected much, whereas in W. A review of damage identification methods that
case of floor beam damage the adjacent two floor responses examine changes in dynamic properties. Shock and
are more sensitive than that of the damaged floor. In vibration digest, 30:95–105, 1998.
addition, it is observed that the ARX model coefficient J. V. A. Dos Santos, H. M. Reis Lopes, and N. M. M.
is sensitive towards characterizing damages between beam Maia. A damage localisation method based on higher
and column irrespective to the damage location. In case of order spatial derivatives of displacement and rotation
first floor beam damage, the difference in ARX model coef- fields. Journal of Physics - Conference Series, 305:1–
ficient fails to perform satisfactorily in characterizing dam- 10, 2011.
age in the beam. If modal and feature-based approaches J. He and Z. F. Fu. Modal Analysis. Butterworth-
are applied together for damage characterization, the na- Heinemann, 2001.

979
2014 ACODS
March 13-15, 2014. Kanpur, India

G. Hedge and R. Sinha. Method of modal identifica- Processing, 23(6):1846–1854, 2009.

tion of torsionally-coupled buildings using eqrthquake M. Novak and L. E. Hifnawy. Damping of structures due to
responses. In The Proceedings of 14th World Conference soil-structure interaction. Journal of Wind Engineering
on Earthquake Engineering, Beijing, China, 2008. and Industrial Aerodynamics, 11(1-3):295–306, 1983.
G. Heo, G. Lee, D. Lee, C. O. Lee, and M. G. Kim. An A. K. Pandey, M. Biswas, and M. M. Samman. Damage
analysis of the characteristics of spg bridges using NExT detection from changes in curvature mode shapes. Jour-
and ERA for real-time monitoring. Key Engineering nal of Sound and Vibration, 145(2):321–332, 1991.
Materials, 270–273:2012–2017, 2004. S. T. Quek, W. Wang, and C. Koh. System identification
G. H. James, T. G. Carne, and J. P. Lauffer. of linear MDoF structures under ambient excitation.
The natural excitation technique for modal Earthquake Enginerring and Structural Dynamics, 28
parameter extraction from operating wind (1):61–77, 1999.
turbines. Technical report, Sandia National S. Ray-Chaudhuri and R. Villaverde. Effect of building
Laboratories, Sandia (prod.sandia.gov/techlib/access- nonlinearity on seismic response of nonstructural com-
control.cgi/1992/921666.pdf), 1993. ponents - A parametric study. Journal of Structural
J. N. Juang and R. S. Pappa. An eigen system realization Engineering, ASCE, 134(4):661–670, 2008.
algorithm for modal parameter identification and modal E. Reinoso and E. Miranda. Estimation of floor accelera-
reduction. Journal of Guidance, 8(5):620–627, 1985. tion demands in high-rise buildings during earthquakes.
J. N. Juang and M. Q. Phan. Identification and control of The Structural Design of Tall And Special Buildings, 14
mechanical systems. Cambridge University Press, 2001. (2):107–130, 2005.
A. Katunin. Damage identification in composite plates K. Roy and S. Ray-Chaudhuri. Autoregressive model for
using two-dimensional b-spline wavelets. Mechanical structural condition assessment in presence of paramet-
Systems and Signal Processing, 25(8):3153–3167, 2011. ric uncertainty. In The Proceedings of International
G. Kerschen, K. Worden, A. F. Vakakis, and J. C. Golinval. Symposium on Engineering under Uncertainty: Safety
Past, present and future of nonlinear system identifica- Assessment and Management. BESU, Shibpur, India -
tion in structural dynamics. Mechanical Systems and 711103, January 3-5, 2012.
Signal Processing, 20(3):505–592, 2006. K. Roy and S. Ray-Chaudhuri. Fundamental mode shape
K. Y. Koo, S. H. Sung, J. W. Park, and H. J. Jung. and its derivatives in structural damage localization.
Damage detection of shear buildings using deflections Journal of Sound and Vibration, 332(21):5584–5593,
obtained by modal flexibility. Smart Materials and 2013.
Structures, 19(11):1–10, 2010. A. Rytter. Vibration based inspection of civil engineering
E. Lourens, C. Papadimitriou, S. Gillijns, E. Reynders, structures. PhD thesis, Department of Building Tech-
G. De Roeck, and G. Lombaert. Joint input-response nology and Structural Engineering, Aalborg University,
estimation for structural systems based on reduced- Denmark, 1993.
order models and vibration data from a limited number P. R. Santa-Ana and E. Miranda. Strength reduction
of sensors. Mechanical Systems and Signal Processing, factors for multi-degree-of-freedom systems. In The
29:310–327, 2012. Proceedings of 12th World Conference on Earthquake
Y. Lu and F. Gao. A novel time-domain auto-regressive Engineering. New Zealand Society for Earthquake En-
model for structural damage diagnosis. Journal of Sound gineering, Auckland, New Zealand, 1997.
and Vibration, 283(3-5):1031–1049, 2005. F. Shen, M. Zheng, D. F. Shi, and F. Xu. Using the cross-
N. M. M. Maia, J. M. M. Silva, and E. A. M. Almas. correlation technique to extract modal parameters on
Damage detection in structures - from mode shapes response-only data. Journal of sound and vibration, 259
to frequency response function methods. Mechanical (5):1163–1179, 2003.
Systems and Signal Processing, 17(3):489–498, 2003. H. Sohn and C. R. Farrar. Damage diagnosis using time
A. Marec, J. H. Thomas, and R. El. Guerjouma. Damage series analysis of vibration signals. Smart materials and
characterization of polymer-based composite materials: structures, 10:1–6, 2001.
Multivariable analysis and wavelet transform for clus- T. M. Whalen. The behavior of higher order mode shape
tering acoustic emission data. Mechanical Systems and derivatives in damaged, beam-like structures. Journal
Signal Processing, 22(6):1441–1464, 2008. of Sound and Vibration, 309(3-5):426–464, 2008.
L. Miller. Non-structural controllability of linear elastic T. Y. Wu, J. C. Chen, and C. C. Wang. Characterization
systems with structural damping. Journal of Functional of gear faults in variable rotating speed using hilbert-
Analysis, 236(2):592–608, 2006. huang transform and instantaneous dimensionless fre-
E. Miranda and S. Taghavi. Approximate floor accelera- quency normalization. Mechanical Systems and Signal
tion demands in multistorey buildings i - formulation. Processing, 30:103–122, 2012.
Journal of Structural Engineering, ASCE, 131(2):203– Y. J. Yan, L. Cheng, Z. Y. Wu, and L. H. Yam. Devel-
211, 2005. opment in vibration-based structural damage detection
B. Moaveni, X. He, J. P. Conte, J. I. Restrepo, and technique. Mechanical Systems and Signal Processing,
P. Marios. System identification study of a 7-story full- 21(5):2198–2211, 2007.
scale building slice tested on the UCSD-NEES shake P. Yuan, Z. Wu, and X. Ma. Estimated mass and
table. Journal of structural engineering, ASCE, 37(6): stiffness matrices of shear building from modal test data.
705–707, 2011. Earthquake Engineering and Structural Dynamics, 27
D. Montalvao, A. M. R. Ribeiro, and J. Duarte-Silva. (5):415–421, 1998.
A method for the localization of damage in a CFRP
plate using damping. Mechanical Systems and Signal

980