Original Article

Structural Health Monitoring

2019, Vol. 18(5-6) 2004–2019

Vibration-based dual-criteria approach Ó The Author(s) 2018

Article reuse guidelines:
sagepub.com/journals-permissions
for damage detection in arch bridges DOI: 10.1177/1475921718810011
journals.sagepub.com/home/shm

Nirmani Jayasundara1 , David Thambiratnam1,

Tommy Chan1 and Andy Nguyen2

Abstract
Vibration characteristics of a structure can be used as an indication of its state of structural health as they vary if the
structural health is affected by damage. This is the broad principle used in structural health monitoring for vibration-
based damage detection of structures. Although most structures are built to have a long life span, they can incur damage
due to many reasons. Early damage detection and appropriate retrofitting will enable the continued safe and efficient
functioning of structures. This study develops and applies a dual-criteria method based on vibration characteristics to
detect and locate damage in arch bridges. Steel arch bridges are one of the most aesthetically pleasing bridge types,
which are reasonably popular in Australia and elsewhere. They exhibit three-dimensional and somewhat complex vibra-
tion characteristics that may not be suitable for traditional vibration-based damage detection methods. There have been
relatively fewer studies on damage detection in these bridge types, and in particular the arch rib and struts, which are
important structural components, have received little attention for damage detection. This study will address this
research gap and treat the damage detection in the arch bridge structural components using the dual-criteria method to
give unambiguous results. The proposed method is first validated by experimental data obtained from testing of a labora-
tory arch bridge model. The experimental results are also used to validate the modelling techniques and this is followed
by damage detection studies on this bridge model as well as on a full-scale long-span arch bridge. Results demonstrate
that the proposed dual-criteria method based on the two damage indices can detect and locate damage in the arch rib
and vertical columns of deck-type arch bridges with considerable accuracy under a range of damage scenarios using only
a few of the early modes of vibration.

Keywords
Structural health monitoring, arch bridge, vibration-based damage detection, modified modal flexibility method, modified
modal strain energy method, noise

Introduction damage detection (VBDD), one of the key aspects of

SHM. It is defined by Chan et al.2 as the use of an on-
Bridges are essential components of a road and trans- structure sensing system to monitor the structural per-
port network and are indispensable for the smooth formance and to evaluate the symptoms of anomalies,
functioning of a city and its economic wellbeing. Most deterioration or damages that may affect the operation,
of the bridges in Australia are ageing and need to be serviceability, or safety and reliability of a structure.
monitored to ensure that they are capable of accommo- Although structures, especially bridge structures, are
dating the current transportation needs with increased designed to have long life spans, they can incur damage
loads and faster speeds of vehicles. In this context,
structural health monitoring (SHM) has recently
emerged as a viable technology to enable the safe oper- 1
Queensland University of Technology, Brisbane, QLD, Australia
ation of both existing and newly built bridges and other 2
University of Southern Queensland, Springfield, QLD, Australia
structures, and the recent Australian Bridge design
Code, AS5100,1 refers to the need for SHM in bridges. Corresponding author:
Nirmani Jayasundara, Queensland University of Technology, 2 George St,
Vibration characteristics of a structure vary if it incurs Brisbane, QLD 4001, Australia.
damage and this is the principle of vibration-based Email: [email protected]
Jayasundara et al. 2005

due to structural deterioration, environmental effects of the structure. Natural frequency has been the para-
and random actions such as impacts. Therefore, dam- meter used in one of the common approaches as it can
age detection prior to unexpected incidents or costly be easily measured from just a few accessible points
repairs has attracted much attention over the years. and it is less contaminated by experimental noise.19 A
There has been considerable research on damage detec- systematic approach for damage detection using mode
tion in simple and complex structures which include shape data was presented by Allemang20 and Lieven
beams,3,4 plate elements,5,6 trusses,7–9 offshore plat- and Ewins21 using modal assurance criterion (MAC)
forms,10,11 bridges12–15 and rail-track structures.16 and coordinate modal assurance criterion (COMAC),
Precise damage detection is one of the key elements respectively, both of which are favourable methods for
of SHM. Two basic types of damage detection can locating damage.
be identified as local methods and global methods.17 The modal flexibility (MF) method, first proposed
Non-destructive tests such as ultrasonic, eddy current, by Pandey and Biswas,22,23 has been used in many
acoustic emission, radiography and magnetic particle damage detection studies due to its accuracy, ease of
inspection are local damage detection techniques which application and convenient computation.5,24 Since the
require prior knowledge of the damaged region.12 The structural MF converges promptly with increasing fre-
limitations of local damage detection techniques were quency, few lower natural frequencies and mass nor-
addressed in global methods which examine the changes malized mode shape vectors can be advantageously
in vibration properties between the healthy and dam- utilized in computing modal flexibilities.22 This method
aged states of the structure to evaluate the damage. has been successfully applied in a wide range of SHM
These methods are known as vibration-based damage and damage detection cases.5,23,25,26
detection techniques (VBDDTs). Comprehensive litera- Modal strain energy (MSE) method as another
ture on VBDDTs provides evidence of the broad VBDDT was first proposed by Stubbs et al.27 This
research carried out in this field over the past few method has been then used by many researches utiliz-
decades. Damage indices (DIs) based on vibration char- ing different measured data for different types of struc-
acteristics are relatively easy to calculate, quick and tures.8,9 A multi-criteria approach incorporating MF
straightforward and have been widely used to detect, and MSE methods was proposed by Shih et al.5 for
locate and quantify the damages in many structures or damage detection in beams and slabs in which they
structural components.18 were able to detect single and multi-damages. It was
Arch bridges are aesthetically pleasing structures shown that the MSE method is capable of detecting
that have been used across the world, including single as well as multiple damages, while the MF
Australia. Out of the many studies available in the liter- method is only good for single damage cases. However,
ature on the damage detection of bridges, there are far the application of these methods for damage detection
less on the damage detection of arch bridges compared in arch bridges is not evident in the literature. Arch
to other types of bridges. The bridge deck in an arch bridges exhibit three-dimensional (3D) and somewhat
bridge has received some attention in previous VBDD complex vibration characteristics which involve the
studies compared to the more critical load-bearing deck, rib and the struts (and as demonstrated later) are
members such as the arch rib, hangers and struts (col- not very favourable for traditional VBDD methods.
umns) of arch bridges. The bridge deck is, however, the This article develops and applies a dual-criteria
most visible component in the bridge, and deck damage approach which simultaneously uses DIs based on
would be more easily captured by the bridge inspectors modified forms of the MF and MSE methods to pro-
than any damage in the other bridge components. vide unambiguous results for detecting and locating
From a structural engineering point of view, all the damage in the main structural components of deck-
other structural components are equally or more type arch bridges. The superior performance of the pro-
important for the safe operation of the bridge. The fail- posed method compared to traditional methods is
ure of one cable/hanger or a strut/column will not be demonstrated through the comparison of the results.
easily visible and can eventually lead to the collapse of Using the two DIs simultaneously enables the results
at least a part of the bridge and subsequently cause the obtained from either DI to complement and supple-
progressive collapse of the whole bridge. This research ment the results from the other DI and lead to more
will therefore focus on developing and applying reliable reliable prediction of the damage location. Prior to its
vibration-based DIs that can provide unambiguous application, the proposed method is validated through
results for detecting and locating damage in arch bridge experimental testing of an arch bridge model under
structural components. laboratory conditions with limited number of sensors.
The vibration-based DI method recognizes struc- The feasibility of the proposed method is demonstrated
tural damage by considering the changes in the vibra- through its application to a range of damage detection
tion properties between the healthy and damaged states scenarios.
2006 Structural Health Monitoring 18(5-6)

Method where k is the bending stiffness of the beam (i.e. the

product of its Young’s modulus E and the second
A dual-criteria approach using two different DIs is
moment of area I) and Fi (x) is the mode shape of ith
developed and applied to detect and locate damage in
modal vector.
arch bridges. These indices are modified versions of the
If the beam is subdivided in to N elements, the MSE
traditional DIs based on the MF and MSE methods,
associated with the jth element or the contribution of
which are briefly presented below.
jth element to the total MSE is Cij, which is given by
equation (5), where kj is the bending stiffness of the jth
MF method element, which is equal to (EI)j
ð
MF F at a location j of a linear structure can be 1 2
expressed as in equation (1) Cij = kj ½f00 i ð xÞ dx ð5Þ
2
j
Xm
1
½F  = f fT ð1Þ For a prismatic member, EI remains constant and
v 2 i i
i=1 i hence kj in equation (5) is equal to k in equation (4).
The fraction of the MSE for the ith mode that is
where i is the mode number considered, and vi and m concentrated in the jth element is given by
are the natural frequency of the structure at mode i and Fij and Fij in equation (6), in its healthy and damaged
the total number of modes considered, respectively. fi states, respectively. In this equation and in what fol-
is the ith mode shape vector and fTi is its transpose. lows, the modal parameters associated with damaged
According to Pandey and Biswas,22 the modal flexi- element are denoted by asterisks
bility change (MFC) can be expressed as in equation
(2), where d and h denote the damage and healthy con- Cij Cij
ditions, respectively Fij = and Fij =  ð6Þ
Ui Ui
" # " #
Xm
1 X m
1 where Cij and Ui are given by
MFC = ½ F d  ½ F h = f fT
2 i i
 f fT
v
i=1 i
v2 i i
i=1 i
d h ð ðL
1   00 2 1   00 2
ð2Þ Cij = 
kj fi  ð xÞ dx and Ui = k fi  ð xÞ dx
2 2
j 0
The diagonal values of MFC matrix are extracted
and each term represents the MF at a particular loca- ð7Þ
tion along the member. In the above equations kj and k  denote the flexural
Modal Flexibility Damage Index (MFDI) is stiffness of the damaged element and the damaged
obtained by dividing MFC value of a particular loca- beam (member), respectively.
tion by MF value extracted from [F] of that same loca- When the beam is subdivided into large number of
tion at healthy state. Therefore, the normalized DI at elements, the fraction of MSE concentrated in each ele-
location j at ith mode can be written as in equation (3) ment is very small, such that for any mode i, the term
hP i hP i Fij and Fij have the following properties
m m
1
i = 1 v2i fi fTi  1
f fT
i = 1 v2i i i
MFDIij = hP D i H
ð3Þ X
N X
N
m 1
f f
i = 1 v2 i i
T
Fij = Fij = 1; and Fij  1, Fij  1 ð8Þ
i H
j=1 j=1

where N is the number of elements in a member.

MSE method If the damage is assumed to be located at a single
MSE-based method proposed by Stubbs et al.27 has subdivision, the fractional strain energy of that element
been improved to suit the damage detection in arch remains relatively constant resulting in
bridges.
For a general Euler–Bernoulli beam, the ith MSE Fij ffi Fij ð9Þ
can be given by equation (4)
If the flexural rigidity (EI) is essentially constant
over the entire length of the member for both the dam-
ðL
1 2 aged and undamaged modes, k of Ui and k* of Ui are
Ui = k ½f00 i ð xÞ dx ð4Þ assumed to be equal.28 Therefore, equation (9) can be
2
0 rearranged so that the ratio of the stiffness kj of the
Jayasundara et al. 2007

Table 1. Ratios of effective mass fractions of the first four Therefore, instead of using the resultant mode
modes of Cold Spring Canyon bridge. shapes as in the traditional DI methods, decomposed
mode shapes using the lateral and vertical components
Mode number Ratios of effective mass to total mass
are used to create the modified DIs used in this study.
X (lateral) Y (vertical) Z To facilitate the use of these component-specific fea-
tures, the DIs are modified as follows
01 0.5117 0.5857E204 0.3323E209
02 0.2066 0.1893E203 0.4094E208 hP i hP i
m m
03 0.1658E209 0.56455 0.1076E208 1
i = 1 v2i fi fTi  1
i = 1 v2i fi fTi
04 0.5040 0.1650E204 0.1875 hP DV i HV
MFDIijV = ð14Þ
m 1
i = 1 v2i fi fTi
HV
damaged element to its stiffness kj in the undamaged hP i hP i
m m
state is obtained as follows
1
i = 1 v2i fi fTi  1
i = 1 v2i fi fTi
hP DL i HL
MFDIijL = ð15Þ
ÐL h i2 m
Ð  00 2 00

1
i = 1 v2i fi fTi
j fi  ð xÞ dx= 0 fi ð xÞ dx
HL
kj fij
= Ð   Ð   = ð10Þ
kj 00 2 L 00 2 fij The above expressions define the modified DIs based
j fi ð xÞ dx= 0 fi ð xÞ dx
on MF where the subscripts L and V denote the lateral
To avoid the possible singularity problems with the and vertical component–specific DIs using and lateral
fij =fij , shifting the axis of reference to avoid numerical and vertical components of mode shapes, respectively.
sensitivities is recommended by Stubbs and Garcia.29 The MSE-based method can be similarly modified
Therefore fij and fij are now considered as 1 + fij and and bijV and bijL are the vertical and lateral
1 + fij , respectively. component–specific DIs using the lateral and vertical
The damage indicator bij is given by components of mode shapes, respectively
h 2 P  00 2 ihP  0 0 2 i
1 + fij
bij = ð11Þ kj f00iV  + fiV  f iV
1 + fij bijV =  = h  2 P 0 0 2 P 00 2 i
  i h  ð16Þ
kj f0 0iV + f iV fiV 
Substituting equation (10) in to equation (11) forms
h 2 P  00 2 ihP  0 0 2 i
Ð ÐL Ð 00
00  2 00  2 L 00 2
kj f  + fiL  f iL
j ½fii ð xÞ dx + 0 ½fii ð xÞ dx
½fi ð xÞ dx iL
kj bijL =  = h ihP  2 i ð17Þ
0
bij = = Ð Ð Ð kj  2 P  2

kj 00 2 L 00 2 L 00  2 f0 0iL + f0 0iL f00iL 
j ½fi ð xÞ dx + 0 ½fi ð xÞ dx 0 ½fii ð xÞ dx

ð12Þ For both methods, a single indicator is generated by

Numerically, it can be expressed as follows taking several global modes into account as follows
h P 00 2 ihP 00 2 i PNM
00  2
kj ðf Þ + fi fi i = 1 Numji
ii bj or DIj = PNM ð18Þ
bij =  = h P 00 2 P 00  2 i
i h ð13Þ
kj 2 i = 1 Denomji
ðfi 00 Þ + ðfi Þ ðfii Þ
In the above equation, NM refers to the number of
modes, and Numji and Denomji are the numerator and
Modified DIs versus traditional DIs denominator of any one the equations (14), (15), (16)
Arch bridges exhibit 3D and rather complex vibration. or (17), depending on the particular DI.
But, the initial global modes of vibration of the whole To illustrate the performance of the component-
arch bridge are governed by the mode shapes of the specific DIs based on both MF and MSE, two cases of
arch rib with dominant contributions in the lateral and arch rib damage in the 213 m span Cold Spring
vertical directions, in which the maximum mass partici- Canyon Bridge (described later in section ‘Damage
pations occur. Table 1 presents the ratios of effective detection in full-scale two hinged arch bridge’) are
mass to total mass fractions of the first four modes of treated. In Figure 1(a), results for the (lateral and verti-
Cold Canyon Bridge extracted from its finite element cal) component-specific DIs based on MSE for rib
(FE) model. It can be seen that modes 1 and 2 in the lat- damage at 250 m are compared. It is clearly evident
eral and mode 3 in the vertical direction are dominant that the lateral component–specific DI performs much
modes with high mass participation. Highest modal better than its vertical counterpart. Figure 1(b) com-
mass participation fraction of each mode is bolted at pares the results from both component-specific DIs
the Table 1. based on MF for damage in the arch rib at 180 m. It
2008 Structural Health Monitoring 18(5-6)

Figure 1. Lateral and vertical (a) MSE and (b) MFDI curves of Cold Spring Canyon Bridge (section ‘Damage detection in full-scale
two hinged arch bridge’) under two cases of rib damage.

this case, the vertical component–specific DI performs Results in Figure 2 clearly show the superior perfor-
better than its horizontal counterpart. mance of the selected component-specific DIs com-
It is therefore evident that depending on the location pared to those of the traditional DIs based on both
of the damage, one of the component-specific DIs, MSE and MF. The enhanced ability of the proposed
based on either MSE or MF method, can perform bet- method to detect damage compared to the traditional
ter than the other. The numerical value of this compo- methods is clearly evident as there is less ambiguity, no
nent DI will be higher compared with that of the other false alarms and there is a more definite indication of
component DI as illustrated in Figure 1. the damage location. Although results from either one
To obtain the best possible results, it is hence neces- of the selected component-specific DI based on either
sary to select (and use) the component-specific DI MF or MSE would give unambiguous results, as seen
which performs better than its counterpart, for both in Figure 2; this study recommends that results from
MF- and MSE-based DIs. Towards this end, the above both DIs be used as they can complement and supple-
MF and MSE methods can be further modified as ment each other and provide adequate confidence in
shown in equations (19) and (20). the predictions.
For the MSE case, the better-performing DI is The method proposed in this article therefore has
selected by comparing the results of bjV and bjH . The three parts:
same is done with the two-component-specific MFDIs.
That is for damage detection at any location, the pro- 1. Use the modal data to calculate the four-compo-
minent bj and DI are obtained by selecting the larger nent-specific (or modified) DIs based on MF and
of the two-component-specific DIs MSE;
     2. Use the above selection criteria (equations (19)–
bj = max bjV , bjH  ð19Þ (21)) to determine the better-performing modified
DIs of both types (i.e. based on MF and MSE);
DI = ½maxðjDIV j, jDIH jÞ ð20Þ
3. Plot the results from both selected modified DIs to
The selected bj and DI are then normalized as shown obtain unambiguous predictions.
 and DI are the mean values
in equation (21), in which b
and sb and sDI are standard deviations of bj and DI, The entire procedure can be automated to make it
respectively easier to use.


bj  b DI  DI
Zj = or Zj = ð21Þ Validation of proposed method and
sb sDI
modelling techniques
To illustrate this choice and compare the perfor-
mance of the selected component-specific DIs with
Experimental testing
those from the traditional DIs, two arch rib damage Before the application of the developed method for
cases involving damage at 205 and 220 m in the same damage detection, a comprehensive laboratory test was
Cold Spring Canyon Arch Bridge are considered. conducted on a small-scale deck-type steel arch bridge
Jayasundara et al. 2009

Figure 2. Comparison of (a) traditional and modified MFDIs and (b) traditional and modified MSE DIs for damage detection in
213 m span Cold Canyon Bridge.

Figure 3. (a) and (b) Accelerometer arrangement (c) random tapping for dynamic test on the bridge.

structure in order to validate the proposed method and bridge was fixed to a plate at the base to represent
the FE modelling techniques. fixed-fixed boundary conditions.
The test model had a span of 1 m and a height of Free vibration testing was performed on the bridge
0.4 m. The deck had a width of 0.25 m and was made model (Figure 3(a)) to obtain the vibration parameters
from 2-mm-thick steel plate, supported by nine, 6-mm- of mode shapes and natural frequencies to validate the
diameter vertical struts at the edges as shown in Figure proposed damage detection method as well as the FE
3. All members of the bridge were made of general steel model. In order to measure lateral and vertical accelera-
with Young’s modulus, density and Poisson ratio of tions, the data acquisition system contained a total
2.05 3 1011 Pa, 7870 kg/m3 and 0.3, respectively. The of 20 single-axial PCBÒ 393B05 integrated circuit
2010 Structural Health Monitoring 18(5-6)

Figure 4. Column damage on experimental model: (a) damage location and (b) accelerometer configuration.

piezoelectric accelerometers which can be self- Experimental validation of the proposed method
calibrated within a few seconds and automatically pick Experimental validation of the developed method was
up the precise acceleration at the position. carried out using the results from the experimental test-
Accelerometers were attached to selected nodes in the ing of the arch bridge model (as described above).
bridge as shown in Figure 3, and each of these will indi- Measured mode shapes and natural frequencies from
cate the acceleration in the direction of its axis. the experimental-free vibration testing of both the dam-
The acceleration data were acquired by a centralized aged and undamaged structure, using a limited number
National Instruments (NI) data acquisition system of sensors and in a typical laboratory environment
including NI cDAQ 9172 chassis, NI 9234 dynamic sig- (which invariably has noise pollution), are used to cal-
nal acquisition modules and an in-house LabVIEW- culate the four-component-specific DIs and then select
based data acquisition program to enable precise the better-performing DIs using the procedure outlined
hardware-based synchronization.30 Since the laboratory through equations (19)–(21). These preferred DIs are
model cannot be excited naturally, random hammer then used to obtain the damage detection results and
tapping was conducted at many different locations/ thereby verify the proposed method experimentally.
directions to simulate traffic and wind excitation on the Accelerometers were attached to the ends of each alter-
bridge (Figure 3(c)). nate column (Figure 4(b)) and cubic spline interpola-
Data were acquired for a predetermined undisrupted tion was used, where necessary, to enhance the mode
duration of 1.5 min so that the continuous signal length shape data.
can be 1000–2000 times of the fundamental period of It is evident that the damage at the base of the third
the bridge (approximate 0.005 s) to enable proper vertical column (C3) of the bridge (Figure 4(a)) is cor-
output-only modal analysis.31 Few repeat tests were rectly detected through the method developed in this
performed to eliminate any random errors and to article by the two preferred component- specific (MF
improve the accuracy of results. The acceleration data and MSE based) DIs as shown in the plots in Figure 5.
were captured in the time domain and was conveyed to The peaks of both modified DIs clearly and correctly
the ARTeMIS modal analysis software to retrieve indicate the location of damage.
modal parameters. The natural frequencies and mode As mentioned earlier, use of either one of the modi-
shapes were determined by the Data Driven Stochastic fied DIs would have sufficed, but results from both are
Subspace Identification (SSI-DATA) in ARTeMIS recommended as they complement and supplement
modal analysis software.32 each other and provide adequate confidence in the
The free vibration test on the undamaged structure predictions.
was considered as the baseline dynamic test which was
followed by a dynamic test performed on the damaged
structure. Physical damage was induced on the struc- Validation of modelling techniques
ture by removing some material from the third vertical The modelling techniques are next validated by com-
column and decreasing its connectivity with the rib, as paring the results from free vibration analysis of the FE
shown in Figure 4(a). The natural frequencies and model of the bridge with those from the experimental
mode shapes obtained from the structure in its healthy testing of the physical bridge (as described above) in
and damaged states will be used to validate the pro- both its healthy and damaged states.
posed method in the next section and the modelling The exact FE model of the laboratory bridge model
techniques in the subsequent section. was created in ANSYS Workbench FE software which
Jayasundara et al. 2011

Figure 5. Predicting third column (C3) damage by (a) modified MFDI and (b) modified MSE DI.

Figure 6. Column damage on (a) experimental model (b) numerical model.

is a commercially available FEM software, capable of experimental model. This was conducted manually by
developing complex structures with multibody parts slightly adjusting the material properties, boundary
and complex analyses. The geometry of the laboratory condition, the fixity between deck plate, frame and
bridge was modelled at ANSYS Workbench other structural components. The masses of the acceler-
DesignModeler as a 3D FE model. Each part (deck, ometers were added at the appropriate location in the
vertical struts, arch rib and wind bracings) was con- FE model. Natural frequencies of the first four modes
nected to the other by using the joint feature in obtained from the FE model in its healthy and dam-
Mechanical module of ANSYS workbench. General aged states are compared with those from the experi-
mild steel properties, mentioned above, were assigned ments in Table 2, while the experimental and numerical
to each element of the structure. All elements, except mode shapes of the bridge model in its healthy state are
the deck elements, were modelled with beam elements shown in Table 3.
(BEAM 188) and the deck was modelled with shell ele- It can be seen that the results from the analysis of
ments (SHELL 181). Additional point masses were the FE model of the laboratory bridge compare reason-
added to the model to account the masses of acceler- ably well with the experimental results and provide con-
ometers. Damage was simulated in the FE model to fidence in the modelling techniques. These modelling
match the physical damage in the bridge structure as techniques are then used to model the full-scale bridge
shown in Figure 6. Prestressed modal analysis was con- that will be used in further damage detection studies.
ducted (incorporating the initial stress state of the
structure under its self-weight) to obtain natural fre-
Damage detection in full-scale two hinged arch
quencies and mode shapes of the numerical model of
the arch bridge.33 bridge
Model updating was performed to tune the struc- To illustrate the applicability of the proposed damage
tural parameters of the FE model to match with the detection technique to a full-scale long-span arch
2012 Structural Health Monitoring 18(5-6)

Table 2. Comparison of natural frequencies from experiments and numerical models under healthy and damaged states of the
structure.

Mode Undamaged state Damaged state

Natural frequencies (Hz) ferror (%) Natural frequencies (Hz) ferror (%)
fexp fFEM fexp fFEM

1 20.00 21.18 5.50 19.00 19.80 4.20

2 33.00 33.36 1.09 33.00 31.85 23.40
3 54.25 53.54 21.30 53.00 50.40 24.50
4 64.25 61.95 23.40 64.00 62.00 20.15

Table 3. Numerical (top) and experimental (bottom) mode shapes of the laboratory scale arch bridge structure.

Mode 1 Mode 2 Mode 3 Mode 4

21.18 Hz 33.36 Hz 53.54 Hz 61.95 Hz

20.00 Hz 33.00 Hz 54.25 Hz 64.25 Hz

Figure 7. Cold Spring Canyon Bridge.

bridge, a complete FE model of Cold Spring Canyon mechanism. The deck is supported over the arch by
Bridge is developed using ANSYS FE modelling soft- vertical columns which are hinged at the panel points
ware. Cold Spring Canyon Bridge is a long-span, deck- and at the end of towers above the arch abutments.
type steel arch bridge with a span of 213 m and a rise The deck slab which connects to the continuing road is
of 36.27 m. The main ribs are restrained except for the restrained longitudinally at one end of the approach
rotational degrees-of-freedom about the transverse axis span. There were no additional expansion joints pres-
at the abutments, thus creating a two hinged arched ent over the arch span34 (Figure 7).
Jayasundara et al. 2013

Table 4. Comparison of mode shapes of Cold Spring Canyon Bridge.

Mode Freq. Present results Results from 2D analysis of Dusseau and Wen34

First mode 0.4564 Hz

Second mode 0.4725 Hz

Third mode 0.826 Hz

Fourth mode 0.849 Hz

2D: two-dimensional.

Table 5. Damage scenarios of arch rib.

Damage scenario

Rib Damage Case 1 Rib Damage Case 2 Rib Damage Case 3

Damaged element Damage at arch spring Damage at 1/4 span of rib Damage at crown
Stiffness reduction 10% 10% 10%

The geometry of the Cold Spring Canyon Bridge demonstrated by its application to a range of different
was developed using Workbench of ANSYS FE soft- damage cases in the Cold Spring Canyon Bridge.
ware. The bridge was modelled as a 3D FE model with Damage was inflicted at three different locations on the
several parts (deck, arch ribs, cross bracings, columns, arch rib and on two different vertical columns of this
etc.) which were ultimately connected via relevant con- bridge as illustrated in Tables 5 and 6. Damage was
nectivity (joint feature) in the mechanical module. The induced as a stiffness reduction by reducing the
global mode shapes of the bridge obtained from the Young’s modulus by 10%. Modal parameters of natu-
present FE model were compared with those from the ral frequencies and mode shapes of the healthy and
two-dimensional (2D) analysis Dusseau and Wen,34 each damaged structure were extracted from FE modal
and reasonably good agreement between the two sets analyses. The proposed modified MF and strain
of mode shapes was obtained, as shown in Table 4. energy–based DIs described in section ‘Method’ are cal-
culated separately for each damage case using the first
Results and discussion four global modes of vibration and are plotted along
the bridge. The peaks in the plots of these DIs are
Single damage scenarios without noise expected to indicate the location of the damage and
The versatility of the proposed dual-criteria method for plots are shown in Figures 8–12 for above five different
damage detection in a long-span arch bridge is damage cases:
2014 Structural Health Monitoring 18(5-6)

Table 6. Damage scenarios of vertical columns.

Damage scenario

Vertical column damage Case 1 Vertical column damage Case 2

Damaged element Damage at mid of the long column Damage at the edge of the short column
Stiffness reduction 10% 10%

Figure 8. Plots of modified MF and modified MSE DIs for rib damage case 1.

Figure 9. Plots of modified MF and modified MSE DIs for rib damage case 2.

 Rib damage case 1: Damage at arch spring  Vertical column damage case 2
 Rib damage case 2: Damage at quarter span of the
arch rib It is clear from Figures 8–12 that both proposed DIs
 Rib damage case 3: Damage at mid span of the arch are capable of detecting damage in the arch rib and ver-
rib tical columns without any false alarms. Although either
 Vertical column damage case 1 one of these modified DI would be effective for damage
Jayasundara et al. 2015

Figure 10. Plots of modified MF and modified MSE DIs for rib damage case 3.

Figure 11. Plots of MMF and MMFDI for damage at middle of long column C1.

Figure 12. Plots of MMF and MMFDI for damage at the edge of short column C5.

detection, this article recommends the simultaneous use to check the accuracy of the proposed method in the
of both DIs to cross check and obtain unambiguous presence of noise in the modal data. Since the vibration
results. responses generated through the FE model are free
from noise, the noise contaminated mode shape data is
created using equation (22) by Shi et al.8
Multiple damage scenarios and effect of noise
 
In the practical context, vibration responses of the ;xi = ;xi 1 + g ux rux j;max, i j ð22Þ
structures are allied with uncertainities in modal fre-
quencies and mode shape data, such as measurement The terms ;xi and ;xi are mode shape component of the
noise and computational errors. Thus, it is important ith mode of vibration at location x with and without
2016 Structural Health Monitoring 18(5-6)

Table 7. Multiple damage scenarios of arch rib and columns with noise polluted data.

Damage case Damage Location Damage Intensity Noise level (%)

Multiple damage case 1 Damage at the rib location 10% stiffness reduction at both locations 15
x = 132.4 m and x = 223.8 m
Multiple damage case 2 Damage at the rib location 5% and 10% stiffness reductions, respectively 15
x = 132.4 m and x = 277.5 m
Multiple damage case 3 Damage at the mid of second 10% stiffness reduction at both columns 10
column and ninth column

Figure 13. Plots of modified DIs for damage case 1 with and without (15%) noise: (a) modified MFDI and (b) modified MSE DI.

Figure 14. Plots of modified DIs with and without 15% noise: (a) modified MFDI and (b) modified MSE DI.

noise, respectively. rux denotes the random noise level the length along the rib and the column number and
and g ux refers to a random number with mean equal to shown in Figures 13–16. The peaks in the plots of these
zero and variance equal to 1 and j;max, i j is the absolute DIs are expected to indicate the location of the damage.
value of the largest component in the ith mode shape. It is clearly evident from these figures that the pro-
Two multiple damage cases of the rib and one multi- posed modified DIs are capable of detecting and locat-
ple column damage case (as illustrated in Table 7) are ing multiple damages in the arch bridge components
considered in this section with and without noise. Each with reasonable accuracy, even in the presence of 15%
case contains similar or different damage intensities. In noise. Furthermore, some traditional DIs are noise
addition, 5%,10% and 15% random noise levels are sensitive and can exhibit false alarms. In such situa-
introduced to mode shapes retrieved through the FE tions the proposed dual-criteria approach provides
analysis. The two DIs are calculated and plotted against the benefit of complementing and supplementing the
Jayasundara et al. 2017

Figure 15. Plots of modified MFDIs with 10% damage at second and ninth columns: (a) without noise and (b) with 10% noise.

Figure 16. Plots of modified MSE DIs with 10% damage at second and ninth columns: (a) without noise and (b) with 10% noise.

Figure 17. Plots of modified MSE and MFDIs with first four and first seven global modes of vibration.

results to provide more accurate predictions of dam- calculated using first seven global modes of vibration.
age location. The results of the two scenarios are illustrated in
Figure 17. A damage of 10% stiffness reduction was
applied at the mid span of the rib and the MMF and
Influence of higher order modes in damage detection MMSE DIs were calculated and plotted using (1) first
The analyses above utilize the first four global modes four and (2) first seven modes of vibration. The results
of vibration to calculate the DIs. In order to check the show that the first four modes are adequate to detect
effect of including higher order modes, the DIs were and locate the damage in the arch rib and the use of
2018 Structural Health Monitoring 18(5-6)

more higher order modes will therefore not contribute Acknowledgements

much towards improving the damage detection. In This research forms a part of a continuing study of SHM of
practice, it is difficult to obtain many accurate higher structures conducted at the Queensland University of
order global modes of vibration of the bridge and Technology (QUT), Australia. The first author gratefully
hence the method proposed in this article is beneficial acknowledges the guidance provided by the supervisors, facili-
as it can detect and locate the damage by using a few ties and support for experiments provided by QUT pilot plant
initial modes of vibration. at Banyo and the QUT Postgraduate Research Scholarship.

Declaration of conflicting interests

Conclusion The author(s) declared no potential conflicts of interest with
Variation in vibration characteristics of a structure is respect to the research, authorship and/or publication of this
article.
used to develop and present a dual-criteria approach
that can accurately detect and locate damage in the
structural components of deck-type arch bridges. In Funding
this approach, the traditional MF- and MSE-based The author(s) received no financial support for the research,
indices are modified by decomposing each of them into authorship and/or publication of this article.
vertical and lateral indices, extracting the larger values
of each type and normalizing them to capture the dam- ORCID iDs
age location very effectively. The proposed method and Nirmani Jayasundara https://orcid.org/0000-0002-7896-
the modelling techniques are experimentally validated 5385
through the testing of a laboratory arch bridge model. Andy Nguyen https://orcid.org/0000-0001-8739-8207
The feasibility of the proposed procedure is illustrated
via its application to detect damage in the major struc-
References
tural components of arch bridges. A range of damage
scenarios are considered in the laboratory bridge model 1. Australia Standards. Bridge design – bridge assessment
(AS 5100.7). Sydney, NSW, Australia: Standards Austra-
and a (full-scale) long-span arch bridge involving dam-
lia, 2017.
age in the arch rib as well as in the vertical columns. 2. Chan TH, Wong K, Li Z, et al. Structural health moni-
Results demonstrate the capability of the proposed toring for long span bridges – Hong Kong experience and
method to detect single as well as multiple damages continuing onto Australia. In: Chan TH and Thambirat-
even in the presence of noise polluted data. Although nam DP (eds) Structural health monitoring in Australia.
both DIs performed reasonably well in all the damage New York: Nova Science, 2011, pp. 1–32.
scenarios treated in this article, this might not be the 3. Rizos P, Aspragathos N and Dimarogonas A. Identification
case in some of the other damage scenarios. The dual- of crack location and magnitude in a cantilever beam from
criteria approach will be very effective in those cases as the vibration modes. J Sound Vib 1990; 138: 381–388.
the results obtained from either DI will complement 4. Hong JC, Kim Y, Lee H, et al. Damage detection using
the Lipschitz exponent estimated by the wavelet trans-
and supplement the results from the other DI and lead
form: applications to vibration modes of a beam. Int J
to more reliable prediction of the damage location.
Solids Struct 2002; 39: 1803–1816.
Reliable prediction of the damage location through the 5. Shih HW, Thambiratnam DP and Chan THT. Vibration
use of the proposed dual-criteria approach will prevent based structural damage detection in flexural members
unsafe decisions or unnecessary examinations of false using multi-criteria approach. J Sound Vib 2009; 323:
alarms. Furthermore, the use of higher modes of vibra- 645–661.
tion for detecting damage in the structural components 6. Cornwell P, Doebling SW and Farrar CR. Application of
was investigated. The results show that for accurate the strain energy damage detection method to plate-like
detection of damage, the first four global modes of structures. J Sound Vib 1999; 224: 359–374.
vibration are sufficient and there are no added benefits 7. Contursi T, Messina A and Williams E. A multiple-
damage location assurance criterion based on natural fre-
by using larger number of mode shapes which are prac-
quency changes. J Vib Control 1998; 4: 619–633.
tically difficult to obtain. Finally, the procedure devel-
8. Shi Z, Law S and Zhang L. Damage localization by
oped in this research can be extended to detect and directly using incomplete mode shapes. J Eng Mech 2000;
locate damage in other types of arch bridges and will 126: 656–660.
contribute towards the safe and efficient performance 9. Law S, Shi Z and Zhang L. Structural damage detection
of these types of bridge structures across Australia and from incomplete and noisy modal test data. J Eng Mech
elsewhere. 1998; 124: 1280–1288.
Jayasundara et al. 2019

10. Wang S, Zhang J, Liu J, et al. Comparative study of 23. Pandey AK and Biswas M. Experimental verification of
modal strain energy based damage localization methods flexibility difference method for locating damage in struc-
for three-dimensional structure. In: The 20th international tures. J Sound Vib 1995; 184: 311–328.
offshore and polar engineering conference, Beijing, China, 24. Praveen Moragaspitiya HN, Thambiratnam DP, Perera
20–25 June 2010, pp.1–6. Mountain View, CA: Interna- NJ, et al. Development of a vibration based method to
tional Society of Offshore and Polar Engineers. update axial shortening of vertical load bearing elements
11. Li H, Yang H and Hu SLJ. Modal strain energy decom- in reinforced concrete buildings. Eng Struct 2013; 46:
position method for damage localization in 3D frame 49–61.
structures. J Eng Mech 2006; 132: 941–951. 25. Toksoy T and Aktan A. Bridge-condition assessment by
12. Shih HW, Thambiratnam D and Chan TH. Damage modal flexibility. Experimental Mechanics 1994; 34:
detection in truss bridges using vibration based multi- 271–278.
criteria approach. Struct Eng Mech 2011; 39: 187–206. 26. Farrar CR and Jauregui DA. Comparative study of dam-
13. Wang FL, Chan TH, Thambiratnam DP, et al. Correla- age identification algorithms applied to a bridge: I experi-
tion-based damage detection for complicated truss ment. Smart Mater Struct 1998; 7: 704–719.
bridges using multi-layer genetic algorithm. Adv Struct 27. Stubbs N, Kim JT and Farrar C. Field verification of a
Eng 2012; 15: 693–706. nondestructive damage localization and severity estima-
14. Shih H, Thambiratnam D and Chan T. Damage detec- tion algorithm. In: Proceedings-SPIE the international
tion in slab-on-girder bridges using vibration characteris- society for optical engineering, Bellingham, WA, 13 Feb-
tics. Struct Control Hlth 2013; 20: 1271–1290. ruary 1995, p.210. Bellingham, WA: SPIE International
15. Wickramasinghe WR, Thambiratnam DP, Chan TH, et Society for Optical.
al. Vibration characteristics and damage detection in a 28. Choi F, Li J, Samali B, et al. Application of the modified
suspension bridge. J Sound Vib 2016; 375: 254–274. damage index method to timber beams. Eng Struct 2008;
16. Jia S and Dhanasekar M. Detection of rail wheel flats 30: 1124–1145.
using wavelet approaches. Struct Hlth Monit 2007; 6: 29. Stubbs N and Garcia G. Application of pattern recogni-
121–131. tion to damage localization. Comput-Aided Civ Inf 1996;
17. Doebling SW, Farrar CR, Prime MB, et al. Damage 11: 395–409.
identification and health monitoring of structural and 30. Nguyen A, Chan TH and Thambiratnam DP. Output-
mechanical systems from changes in their vibration char- only modal testing and monitoring of civil engineering
acteristics: a literature review. Los Alamos National Lab, structures: instrumentation and test management. In:
Los Alamos, NM, 1 May 1996. Proceedings of the international conference on structural
18. Salehi M, Rad SZ, Ghayour M, et al. A non model-based health monitoring of intelligent infrastructure (SHMII-
damage detection technique using dynamically measured 08), Brisbane, QLD, Australia, 5–8 December 2017,
flexibility matrix. IJST-T Mech Eng 2011; 35: 137–149. pp.1–12. New York: Curran Associates, Inc.
19. Fan W and Qiao P. Vibration-based damage identifica- 31. Nguyen T, Chan TH, Thambiratnam DP, et al. Develop-
tion methods: a review and comparative study. Struct ment of a cost-effective and flexible vibration DAQ sys-
Hlth Monit 2011; 10: 83–111. tem for long-term continuous structural health
20. Allemang RJ. The modal assurance criterion – twenty monitoring. Mech Syst Signal Pr 2015; 64–65: 314–324.
years of use and abuse. Sound Vib 2003; 37: 14–21. 32. Nguyen T, Chan TH and Thambiratnam DP. Effects of
21. Lieven N and Ewins D. Spatial correlation of mode wireless sensor network uncertainties on output-only
shapes, the coordinate modal assurance criterion modal analysis employing merged data of multiple tests.
(COMAC). In: Proceedings of the 6th international modal Adv Struct Eng 2014; 17: 319–330.
analysis conference, Kissimmee, FL, 1–4 February 1988, 33. ANSYSÒ. Academic research, Release 18, help system,
pp.690–695. New York: Union College. coupled field analysis guide. Canonsburg, PA: ANSYS,
22. Pandey AK and Biswas M. Damage detection in struc- Inc., 2017.
tures using changes in flexibility. J Sound Vib 1994; 169: 34. Dusseau RA and Wen RK. Seismic responses of deck-
3–17. type arch bridges. Earthq Eng Struct D 1989; 18: 701–715.