Engineering Structures 99 (2015) 173–183

Contents lists available at ScienceDirect

Engineering Structures
journal homepage: www.elsevier.com/locate/engstruct

Structural identification and damage detection through long-gauge

strain measurements
Jian Zhang ⇑, S.L. Guo, Z.S. Wu, Q.Q. Zhang
Key Laboratory of C&PC Structures of the Ministry of Education, Southeast University, Nanjing 210096, China
International Institute for Urban Systems Engineering, Southeast University, Nanjing 210096, China

a r t i c l e i n f o a b s t r a c t

Article history: Strain measurements from traditional point-type gauges only reveal structural local information. This
Received 2 April 2014 feature significantly limits the development and application of strain modal theory in civil engineering.
Revised 12 April 2015 The long-gauge fiber optic strain sensor emerging recently has the unique merit of reflecting both local
Accepted 13 April 2015
and global information of a structure by measuring the averaged strain within a long gauge length
(e.g., 1–2 m). Based on this advantage, strain modal identification by processing the dynamic macro strain
measurement is investigated. First, the macro strain frequency response function is derived based on the
Keywords:
mapping relationship between the long gauge strain and displacement. Then, two methods are proposed
Long-gauge sensing
Macro strain
for structural strain modal identification and their results are compared. Moreover, damage indexes
Signal processing based on identified macro strain modal shapes are investigated for structural damage detection. The
Strain modal identification example of a steel stringer bridge model successfully illustrates that the proposed methods accurately
Damage detection identify structural macro strain modal shapes, and the results also reveals that macro strain based
Singular value decomposition indexes are much more robust than traditional modal parameter based indexes for structural damage
detection.
Ó 2015 Elsevier Ltd. All rights reserved.

1. Introduction [6]. Hwang et al. developed a procedure to estimate the effective

modal mass of the structure based on the H-infinity optimal model
A number of structural health monitoring (SHM) technologies technique [7]. Zhang and Moon proposed to identify structural
has been developed in civil engineering, and some of them have modal flexibility from multiple reference impact test data [8].
been successfully applied on real-life structures throughout the These acceleration-based methods have their scientific values.
world [1–5]. The SHM systems utilized in those applications gener- However, the identified basic modal parameters for instance fre-
ally consists of many types of sensors. For instance, over 1400 sen- quencies and mode shapes are global indicators which are not sen-
sors including 16 categories were installed on the Sutong Yangtze sitive to minor structure changes for instance local damages. It
River Bridge, the world’s second longest cable-stayed bridge, in tends to that more and more bridge engineers lost their patience
China. The accelerometer is the most widely used sensor among on SHM systems because they think modal identification results
them. Acceleration-based modal identification methods have been from acceleration measurements are too global to directly support
developed over fifty years and they have become the mainstream decision-making of bridge maintenance and management.
to process the ambient vibration data of large-scale structures As an alternative way of measuring structural responses, strain
[5]. Peeters et al. processed the Z24 Bridge’s monitoring data of gauge has the capacity to monitor structural local information. It
one entire year and they confirmed that the identified frequency has been widely used in SHM systems of civil structures for
is influenced by temperature [1]. Melo et al. installed a wireless on-line monitoring and damage detection [9–14]. Shenton et al.
sensor network on the Golden Gate Bridge, from which modal recommended the use of static strain measurements to measure
parameters were identified and statistically analyzed [2]. the dead load redistribution when structural damage occurs [15].
Brownjohn et al. proposed a method to directly estimate the modal Catbas et al. used over 200 strain sensors to monitor the most crit-
mass of a footbridge by vibrations caused by pedestrian excitation ical electrical, mechanical and structural components of a bridge,
and they proposed an approach through tracking correlation coef-
⇑ Corresponding author at: Key Laboratory of C&PC Structures of the Ministry of ficients between strain time histories at different location for dam-
Education, Southeast University, Nanjing 210096, China. age detection [16]. Xu et al. proposed a damage detection strategy
E-mail address: [email protected] (J. Zhang). by defining a strain-based energy index derived from the strain

http://dx.doi.org/10.1016/j.engstruct.2015.04.024
0141-0296/Ó 2015 Elsevier Ltd. All rights reserved.
174 J. Zhang et al. / Engineering Structures 99 (2015) 173–183

FRFs and energy spectral density of strain responses [17]. Phares development of a kind of long-gauge FBG sensor of which the effec-
et al. installed two sacrificial specimens on damage-sensitive loca- tive sensing gauge length can be extended to several centimeters
tions of a bridge then they used the strain data collected from sac- or meters through special design and manufacturing. The scheme
rificial specimens for structural damage identification through a of a packaged long-gauge FBG sensor is shown in Fig. 1(a). Its prin-
statistical algorithm [18]. The above methods have their merits ciple feature is the handling of an embedded tube inside which a
for damage detection but they were developed based on the bare optic fiber with FBG is sleeved and fixed at two ends to ensure
assumption that strain gauges were installed covering at least at the measured value represents the averaged strain over the gauge
the vicinity of the damaged area, otherwise it may cause a failure length. In addition, long-gauge sensors can be connected in series
in damage identification. Traditional strain gauges are point-type to make a FBG sensor network for area macro-strain measuring
sensors while civil structures generally are large-scale and com- (Fig. 1(b)), thus enabling a global monitoring of the structures.
plex. Therefore, it remains a challenging issue for damage detec- Therefore, this novel sensor has the unique feature to measure
tion of civil structures by using point-type strain sensors. not only local but also global information of the structure. It has
Distributed strain sensing technologies for instance the Brillouin been tested and applied in civil structures as a new and interesting
Optical Time Domain Analysis (BOTDA) technology are promising possibility for macro strain measurement.
for civil structures but they have not been widely adopted in engi-
neering practices due to their high cost and limited resolution [17]. 3. Macro-strain modal identification theory
To overcome the shortcomings of accelerometers and
point-type strain gauges described above, the long-gauge strain Development of the long-gauge strain sensor provides a new
sensing technique have been developed recently, which aims at opportunity for developing the macro-strain modal identification
measuring both global and local information of the structure. Wu theory because it reveals not only local but also global information
et al. developed a long-gauge fiber brag grating (FBG) sensor of the structure. The macro strain FRF will be first derived based on
[19,20] to measure the averaged strain of the critical structural ele- the mapping relationship between the long-gauge strain and
ment over a long gauge length (for instance, 1–2 m). This smart displacement.
sensor achieves the expansion of sensing area from point to a long Suppose the beam is an Euler–Bernoulli type with two
gauge length. Moreover, a number of long-gauge FBG sensors can local degree-of-freedoms (one for vertical translation v and the
be connected in series to make a sensor network for distributed other for rotation h) at each node. Under the condition of single
strain measurements. Therefore, the long gauge sensing technique load excitation, the long gauge strain measurement of the element
has the unique feature to measure not only local but also global m is:
information of the structure. Strain modal theory has been devel-
em ¼ lm ðho  hp Þ ð1Þ
oped for a long time [21,22] but its maturity and its application
in engineering practice is far below the acceleration-based modal where lm ¼ hm
, hm is the distance from the beam bottom to the neu-
Lm
identification theory. This is because it is difficult to identify struc-
tral axis of the beam, Lm is the element length, ho and hp are the rota-
tural modal parameters (global property) from the point-type
tions of two nodes of the element. It should be noted that em is the
strain measurement (local information). The unique feature of
averaged strain within the gauge length. The exact strain distribu-
the developed long-gauge strain sensor provides a new opportu-
tion along the element length is not studied. The FRF matrix
nity for developing macro-strain modal theory.
describes the relationship at a measured response point for a mea-
Numerous methods have been developed to process
sured input point as a function of frequency. Based on the definition
acceleration-based data for structural modal identification, but
of the strain frequency response function (SFRF) and Eq. (1), it is
very few methods are available for strain modal identification. In
derived that
this article, the macro-strain modal identification theory corre-
sponding to the long-gauge strain measurements will be devel- ðho ðxÞ  hp ðxÞÞ  
Hemq ðxÞ ¼ lm

oped. The basic idea is developing the mapping relationship ¼ lm Hhoq ðxÞ  Hhpq ðxÞ ð2Þ
f q ð xÞ
between the displacement and long-gauge strain, then using it to
where Hemq ðxÞ denotes the macro strain FRF of the element m due to

identify structural macro-strain modal parameters with a similar
procedure of acceleration-based modal theory. Two methods, one the input force at the node q. Hhoq ðxÞ and Hhpq ðxÞ correspond to the
is based on the characteristic of the derived macro-strain fre- rotational displacement FRFs of the node o and p respectively. Based
quency response function (FRF), another is through the singular on the acceleration-based modal theory, the macro strain FRF is fur-
value decomposition (SVD) of the macro-strain FRF matrix, will ther derived as shown below,
be developed to identify strain modal parameters. Three damage
 
indexes using the identified macro strain mode shapes, including lm /hor  /hpr /vqr
X
N
Hemq ð
the change of the macro-strain mode shape, it is curvature change 
xÞ ¼   ð3Þ
and it is curvature squares change, will be further investigated and r¼1
M r x2r  x2 þ 2jnr xr x
they will be compared with the traditional modal parameter based qffiffiffiffiffi
Kr
damage indexes. The example of a steel stringer bridge model will where xr ¼ Mr
is the r-th modal circular frequency of the struc-
be investigated to illustrate the effectiveness of the proposed ture, K r and Mr are modal stiffness and modal mass respectively,
method for strain modal identification and to demonstrate its nr ¼ 2MCrrxr is the r-th modal damping ratio, /v is the vertical dis-
superiority over the traditional damage indexes for structural dam-
placement mode shape, and /h is the rotational displacement mode
age detection.  
shape. Similar to Eq. (1), it is defined that /emr ¼ lm /hor  /hpr ,


2. Long-gauge FBG sensor where /emr is the macro strain mode shape. Therefore, when the


force is applied at point p, the corresponding long-gauge SFRF at

FBG sensors in nature hold the excellent ability to provide a element m, Hemq ðxÞ, is rewritten as follow:


measurement having higher precision and measuring stability.

But almost all commercially available FBG sensors are still affili- X
N
/emr /vqr


Hemq ðxÞ ¼

ated to point-type gauges as their inherent gauge length is on   ð4Þ
M r x2r  x2 þ 2jnr xr x
the order of a few millimeters. This section will introduce the r¼1
 J. Zhang et al. / Engineering Structures 99 (2015) 173–183 175

Fig. 1. Packaged long-gauge FBG sensor: (a) schematic figure; (b) sensor array.

Eq. (4) represents that the macro strain FRF incorporates the in the q-th column of the strain FRF matrix in Eq. (5), it is seen that
mass, stiffness and damping ratio of the structure. It is the basis /vqr
those elements consist of a constant, C r ðxÞ
. Thus it can be rewritten
of the proposed strain modal identification method described
in the following format:
below.
X
N
/vqr
He:q ðxr Þ¼ /e:r
 
3.1. Strain modal identification method I ð6Þ
r¼1
C r ð xr Þ
The first way proposed to identify strain mode shapes is using
where He:q ðxr Þ denotes the q-th column of He ðxr Þ, and xr is the
 
the characteristic of the derived macro strain FRF. When the h iT
structure is excited at multiple points and the response is mea- structural frequency in the r-th mode. /e:r ¼ /e1r ; . . . ; /epr
  
is the
sured at multiple points, the strain FRF matrix is
r-th strain mode shape. It is known that the FRFs can be decom-
2 3
X
n
 /v posed into mode spaces and the r-th component will be prominent
6 /e1r C r ðqrxÞ 7 when x ¼ xr as shown in Fig. 2. Therefore, it can be supposed that
6 7
6 r¼1 7 He:q ðxr Þ  r He:q ðxr Þ at the r-th frequency xr , where r He:q ðxr Þ is the
  
6 . 7
6 . 7 /vqr
6 . 7 r-th component in the modal space. Because the coefficient
6X n Xn Xn 7 C r ðxr Þ
e 6 /pr
e
v e /qr v /pr
e
v 7
½H ðxÞ ¼ 6
6 /
C r ðxÞ 1r
  / pr C r ðxÞ   / 7
C r ðxÞ N o r 7 ð5Þ in Eq. (6) is a constant relative to the r-th frequency, it can be
6 r¼1 r¼1 r¼1 7 derived that
6 7
6 . 7
6 .. 7
He:q ðxr Þ  r He:q ðxr Þ / /e:r
  
6 7 ð7Þ
6 Xn v 7
4 e / qr 5
/Ni r Cr ðxÞ
The symbol / denotes the linearly proportional relationship. It
r¼1
is seen from Eq. (7) that the strain mode shape can be identified by
where N o and Ni represent the output and input number respec- picking the peaks of the strain FRFs in a column. To get a column of
 
tively. C r ðxÞ ¼ Mr x2r  x2 þ 2jnr xr x . By checking the elements the SFRF matrix, it only needs excite the structure on a single node

Fig. 2. Identification of strain modal parameters from strain FRF features.

176 J. Zhang et al. / Engineering Structures 99 (2015) 173–183

and measure the response of all output nodes, thus it is easy to per- of damage is potential to be detected. The strain MSC index is
form strain modal identification in the above described way. For expressed as follow:
easy understanding, Fig. 2 is plotted to illustrate the way to iden- X 
/emr  /emr

tify structural strain mode shapes by picking the peaks of the strain MSCm ¼ ð12Þ
r
FRFs in a column.

where /emj and /emj are the macro-strain mode shape values of the


3.2. Strain modal identification method II element m after and before the damage.

Singular value decomposition (SVD) technique has been applied 4.2. Strain Mode Shape Curvature Change (MSCC) index
in traditional modal theory to improve the accuracy of the identified
modal parameters from acceleration measurements. It decomposes This index is defined as
an acceleration FRF matrix at a spectral line in the following way: X  00  00
T MSCCm ¼ /emr  /emr ð13Þ
HðxÞ ¼ USV ð8Þ m

where the singular value matrix, S, is a diagonal matrix with real,  00

00  00
/eðmþ1Þr 2/emr þ/eðm1Þr
  00

where /emr ¼ , it is analytical expression of the

non-negative elements in descending order, the U and V matrices L2m

are orthogonal matrices with unity length, i.e., UU ¼ I, VV ¼ I, T T curvatures by numerical differentiation of mode shapes at measure-
which are called the left and right singular vectors, respectively. ment points. Lm is element length. It is believed that for a given
It is known that the acceleration FRF can be written in the fol- moment applied to a structure, a reduction of the stiffness caused
lowing format: by damage induces a raise in curvature.
 
1 4.3. Strain Mode Shape Curvature Square Change (MSCSC) index
Hðxk Þ ¼ / LT ð9Þ
jxk  cr
where xk is the frequency line, k¼1 to Ns , Similarly, the MSCSC index is defined as
2 1
3 X
jxk c1
 0 MSCSCm ¼ /emr
 002
 /emr
 002
ð14Þ
h i 6 7 qffiffiffiffiffiffiffiffiffiffiffiffiffiffi
1
¼6 .. .. .. 7, cr ¼ nr xr þ jxr 1  n2 , and c is r
jxk cr 4 . . . 5 r r
1
0  jxk cN The above three indexes are defined by using the identified
the conjugate of cr . xr is structural frequency, nr is damping ratio, macro strain mode shapes. Their effectiveness for damage detec-
Lr is modal participation factor, / is the displacement mode shape. tion will be studied in next section by comparing with the tradi-
By comparing Eqs. (8) and (9), it is found that S consists of the infor- tional indexes using displacement mode shapes.
mation of structure frequencies and damping ratios, and displace-
ment mode shapes can be extracted from U. 5. Example investigations
Substituting Eq. (8) in Eq. (2), it is derived that
  5.1. Experimental example of a simple supported beam
He ðxÞ ¼ gm U o So V To  U p Sp V Tp

ð10Þ
Multiple-reference impact testing of a simple supported beam
where, So and Sp are same because they consist of the frequency and was performed to verity the effectiveness of the proposed method
damping information of the same structure. V o and V p consists of for long-gauge strain modal identification and damage detection.
the information of modal participation factors. They are corre- The test specimen is a steel beam with a I-shape cross section as
sponding to the same impacting locations of the structure in the shown in Fig. 3. The length of the beam between the support
multiple-reference impact test. Thus they are same too. In contrast,
U o and U p are different because U o consist of information of the
mode shapes of the left nodes of all elements while U p is for the
right nodes of all elements. Therefore, Eq. (10) can be rewritten as,

He ðxÞ ¼ gm ðU o  U p ÞSV T ¼ U e SV T
 
ð11Þ
e
where U ¼ gm ðU o  U p Þ. Eq. (11) represents the results of perform-
ing the singular value decomposition of the macro-strain FRF. It is
seen that strain mode shapes are obtained from the left singular
vector, U e . The identified results from the method II will be more


robust than those from the method I because the SVD technique
is able to reduce the noise effect.

4. Macro mode strain based damage detection

Three damage indexes using the macro strain modal identifica-

tion results are investigated in order to illustrate the usefulness of
the long-gauge strain measurement and related macro strain
modal identification method.

4.1. Strain Mode Shape Change (MSC) index

By graphically overlapping the strain mode shapes of the struc-

ture before and after the damage at a selected mode, the positions Fig. 3. The experimental specimen.
 J. Zhang et al. / Engineering Structures 99 (2015) 173–183 177

A1 A2 A3 A4 A5 A6 A7 A8 A9 A10 A11
10 68 10
10

F1 F2 F3 F4 F5 F6 F7 F8 F9 F10 F11 F12

8 160

68
6 10
1 2 3 4 5 7 8 9 11 12
1 2 3 4 5 6 7 8 9 10 11 12 13 accelerometers
long-gauge
D1˖two damages at elements 6 and 10
FBG sensor

Fig. 4. The experimental sensor layout.

Fig. 5. Typical impacting force, macro strain responses, and strain FRF.

locations is 6 m. The steel material has a density of 20.5 kg/m, and through the proposed method, two status of the structure were
its Young’s Modulus is 206 Gpa. The beam cross section has an area investigated. One is the intact status denoted by D0, another is
of 2:61  103 m2 and its moment of inertia about weak axis is the damaged status denoted by D1. For the damaged structure,
1:13  105 m4 . The test specimen was oriented on two steel ped- the top flange widths of the 6th and 10th elements were reduced
estals so that it would bend about its weak axis under vertical as shown in Fig. 4, which equals reducing 6.6% stiffness of those
loads. The physical structure was excited by hammer impacting elements. 11 accelerometers were also mounted on the simple
on the references nodes. 12 long-gauge fiber optic sensors with supported beam with equal intervals to observe structural acceler-
the gauge length of 0.5 m were employed on the bottom of the ations during the impact test.
simple supported beam with equal intervals. The sampling of the Typical impacting force and long-gauge strain responses are
impacting test data is 0.001 s. To perform damage detection plotted in Fig. 5(a) and (b) respectively for illustration purpose,
and Fig. 5(c) shows a typical long-gauge strain FRF estimated from
-5 the impact forces and strain responses. Two strain modal identifi-
10
cation methods using Eqs. (5) and (8) respectively has been pro-
posed in the second section of this article. The first method
directly using the strain FRFs to estimate structure strain mode
Log Magnitude

-6
10 shapes, while the second method performs single value decompo-
sition of the strain FRF matrix based on the complex modal indica-
tor function (CMIF) method. Because the second method is more
-7 robust for strain modal identification, only the results from the sec-
10
ond method is provided. Fig. 6 illustrates the CMIF plots from the
experimental data, in which the red1 circles on the curve peaks
denote structure modes. The number of curves equals the number
-8
10 of impacting nodes. Structure frequencies in the first three modes
0 50 100 150
Frequency (Hz)
1
For interpretation of color in Figs. 6, 10 and 13, the reader is referred to the web
Fig. 6. CMIF plot of the macro strain FRFs. version of this article.
178 J. Zhang et al. / Engineering Structures 99 (2015) 173–183

D0 D1 D0 D1

(a) (b)

Displacement Mode Shape

1 1

Strain Mode Shape

0.8 0.8

0.6 0.6

0.4 0.4

0.2 0.2

0 0
1 3 5 7 9 11 13 2 4 6 8 10 12
Node Number Element Number

Fig. 7. Identified mode shapes, (a) displacement mode shapes, (b) strain mode shapes.

0.01 0.05
Damage 1ndex (a)

Damage 1ndex (a)

0.008 (a) 0.04 (b)
0.006 0.03
0.004 0.02
0.002 0.01

0 0
1 3 5 7 9 11 13 2 4 6 8 10 12
Node Number Element Number

Fig. 8. The MSC index for damage detection, (a) displacement MSC, (b) strain MSC.

0.08 0.4
Damage 1ndex (b)

Damage 1ndex (b)

0.06 (a) 0.3 (b)

0.04 0.2

0.02 0.1

0 0
1 3 5 7 9 11 13 2 4 6 8 10 12
Node Number Element Number

Fig. 9. The MSCC index for damage detection, (a) displacement MSCC, (b) strain MSCC.

0.04 0.4
Damage 1ndex (c)

Damage 1ndex (c)

0.03 (a) 0.3 (b)

0.02 0.2

0.01 0.1

0 0
1 3 5 7 9 11 13 2 4 6 8 10 12
Node Number Element Number

Fig. 10. The MSCSC index for damage detection, (a) displacement MSCSC, (b) strain MSCSC.

of the structure are identified to be 14.06 Hz, 55.50 Hz and 118.2 Hz structural damage detection as shown in
respectively. Fig. 7(b) shows the identified strain mode shapes of the Figs. 8(b), 9(b) and 10(b) respectively, in which the red triangles
intact and damaged structures respectively. For comparison, the denote exact damage locations. It is seen that all three damage
similar procedure is also performed to the measured acclerations, indexes successfully identified damage locations. In contrast,
and the identified displacement mode shapes are plotted in Fig. 7(a). the similar damage indexes using displacement modal parame-
After long-gauge strain modal parameters are identified, three ters identified from accelerations failed to identify damage loca-
damage indexes proposed in Eqs. 12–14 are calculated for tions as shown in Figs. 8(a), 9(a) and 10(a) respectively.
 J. Zhang et al. / Engineering Structures 99 (2015) 173–183 179

Fig. 11. The steel stringer bridge model, (a) side view; (b) FE model.

4
Force (kN)

-20
(a) (c)
2

Log (|SFRF|)
0 -22
0 1 2 3 4 5
Time (s)
0.02 -24
Strain (με)

(b)
0

-0.02 -26
0 1 2 3 4 5 0 10 20 30
Time (s) Frequency (Hz)

Fig. 12. Typical impacting force, macro strain responses, and strain FRFs.

5.2. Steel stringer bridge example 5.3. Macro strain modal identification results

To verify the effectiveness of the proposed strain modal identi- The typical impacting force applied on the element 15 and cor-
fication method by using the long-gauge strain measurement, a responding long-gauge strain responses are plotted for illustration
steel stringer bridge as shown in Fig. 11(a), considered as the purpose in Fig. 12(a) and (b) respectively. In a similar way to get
benchmark bridge in the International Bridge Study (IBS) initiated FRFs from accelerations, macro strain FRFs were estimated from
by Federal Highway Administration (FHWA) Long-Term Bridge the impacting force and long-gauge strain data (Fig. 12(c)).
Performance Program (LTBP), is taken under examination. The Totally there are 4 impacting nodes and 128 channels of outputs,
investigated bridge has four spans with a standard steel stringer thus the dimension of the estimated macro strain FRF matrix is
design of girders. In each span, the girder is supported by stringers 128 by 4 at a frequency line.
made of I-shaped beams and connected laterally by bracings. The Two methods proposed in Section 3 were used to identify
deck of the bridge was cast using stay-in-place forms. macro-strain modal parameters by processing the strain FRFs. In
In order to verify the effectiveness of the macro mode strain the first method utilizing the feature of the strain FRFs (Eq. (5)),
based indexes for structural damage detection while it is not real- strain mode shapes were identified by picking the peaks of a col-
istic to make damages on the real bridge, long-gauge dynamic umn of strain FRFs. In the second method, the SVD technique
strain measurements are simulated by the finite element analysis. was performed to the strain FRF matrix (Eq. (11)). It produced
the left singular value vector U e of size 128 by 4, the singular value

The second span on the southbound side of the bridge was taken
into consideration during the FE modeling. This span is composed
by eight girders with the skew on one side as shown in Fig. 11(b).
-8
The distance between each girder is 8 ft. Each girder was subdi- 10
vided into 32 elements for FE modeling. Multiple reference impact
test data were simulated by the FE analysis in the Sap2000 soft-
ware. The impacting forces were applied at centers of the elements
15, 53, 75 and 112 of the structure respectively (Fig. 11(b)), and the
Log (SV)

corresponding time history of long-gauge strain responses of each 10

-9

element under each impacting were simulated. 10% of white noise

was added both into the input impacting force and output dynamic
strain responses to simulate the observation noise, where 10%
means that the variance of the added white noise is 10% of the vari-
ance of the simulated data. Even though the responses of all eight -10
10
girders were simulated, only four of them were processed for 0 5 10 15 20 25
demonstration purpose, which were the first, third, sixth and Frequency (Hz)
eighth girder as shown in Fig. 11(b) with a length of 129.25,
122.25, 111.75 and 104.75 ft respectively. Fig. 13. CMIF plot of the macro strain FRFs.
180 J. Zhang et al. / Engineering Structures 99 (2015) 173–183

1 1

MAC

MAC
0.5 0.5

10 10
0 0
8 8
0 0
2 6 2 6
4 4 4 4
6 6
8 2 8 2
10 0 10 0

(a) The method I (b) The method II

Fig. 14. MAC values of the identified results, (a) the method I, (b) the method II.

f1=3.0/3.0Hz f2=3.6/3.6Hz f3=4.3/4.4Hz f4=6.6/6.6Hz f5=9.6/9.6Hz

f6=11.2/11.1Hz f7=12.4/12.4Hz f8=13.8/13.7Hz f9=18.3/18.3Hz f10=23.6/23.8Hz

Fig. 15. Identified strain mode shapes.

Fig. 16. Damaged states of the structure with reducing cross sections.

matrix S sized 4 by 4, and the right singular value vector V of size 4 resonance. The number of curves in Fig. 13 corresponds to the
by 4. Since the left and right singular vectors are unitary matrices, number of inputs in the multiple-reference impact test. As shown
the amplitude information is carried within the singular value in the plot, the peaks of the blue spectral line marked by red circles
matrix. The singular values were plotted as a function of frequency represent structural modes. Namely, structural frequencies were
in Fig. 13. Along the frequency line near a resonance, the system determined from the significant singular values identified at the
pole k and input frequency xk are closer which results in a local maximums of the plot.
maximum of the singular value plot. Thus, the singular value peaks Consequently, the strain mode shapes of the tested structure
were identified from the left singular vector U e through Eq. (11).

are the possible pole locations of the structure. The left singular
vector associated with the singular value peak is the approximate MAC values between the identified strain mode shapes from the
strain modal vector of the system, being the response at the first method and the exact values from Sap2000 software are
 J. Zhang et al. / Engineering Structures 99 (2015) 173–183 181

D0 D1 D0 D1
2 (a) 2 (b)
1.5
1.5

1
1
0.5

0.5
0
Girder1 Girder3 Girder1 Girder3
0 -0.5
1 10 20 30 40 50 60 66 1 10 20 30 40 50 6064
Node Number Element Number

D0 D2 D0 D2
2 (c) 2 (d)
1.5
1.5

1
1
0.5

0.5
0
Girder1 Girder3 Girder1 Girder3
0 -0.5
1 10 20 30 40 50 60 66 1 10 20 30 40 50 6064
Node Number Element Number

Fig. 17. The MSC index for damage detection, (a) displacement MSC for D1 damage state, (b) strain MSC for D1 damage state, (c) displacement MSC for D2 damage state, (d)
strain MSC for D2 damage state.

plotted in Fig. 14(a). Similarly, Fig. 14(b) is for the results from the the first mode. It is seen that the strain mode shapes before and
second method. As can be seen, both methods accurately identified after damages have obvious changes at the damage locations. In
the first 8 modes, while the second method identified higher contrast, the first displacement mode shapes of undamaged and
modes for instance the 9th and 10th modes. This is because the damaged structures plotted in Fig. 17(a) are almost same, which
SVD technique is able to enhance the data quality by reducing fails to identify damage locations. Comparisons between undam-
noise. Both methods have the ability to identify structural modes, aged state D0 and damaged states D12 using the MSC indexes with
while the second one is more robust. Thus only the results from strain and displacement mode shapes are also plotted in
the second methods are provided in the following sections. The Fig. 17(d) and (c) respectively. They illustrate the same findings
identified strain mode shapes in the first 10 modes from the sec- that the strain MSC index is much more effective than the displace-
ond method are illustrated in Fig. 15. ment MSC index for damage detections.
The second damage index, macro strain mode shape curvature
5.4. Macro modal strain based damage detection results change (MSCC) as defined in Eq. (13), was calculated from the iden-
tified strain modal parameters for damage detection. Fig. 18(b) illus-
Two damaged states denoted by D1 and D2 respectively were trates the macro modal strain based MSCC index for the D1 damage
simulated in the bridge model as shown in Fig. 16 to investigate situation. Similarly, Fig. 18(d) is for the D2 damage situation. For
the effectiveness of macro modal strain based indexes for damage comparison, the MSCC index using the displacement mode shapes
detection. In D1 situation, a shortening of 0.12 ft in the flanges of are also provided in Fig. 18(a) and (c) for the damage situations D1
the section of elements 16 and 24 was introduced which caused and D2 respectively. It is clear that strain MSCC index perform well
a reduction of 11.67% in the elemental stiffness. In D2 situation, for damage detection even in case with 10% noise. In contrast, the
other than reducing 11.67% stiffness of elements 16 and 24 on damages are correctly identified by the displacement MSCC index
the first girder, two additional shortenings of 0.13 ft were made in noise-free case, but it fails in the presence of noise even when
to the flanges of elements 40 and 52 on the third girder which the noise is low-level for instance 0.5%. These figures indicate that
induce a reduction of 14.40% in the stiffness of each element. strain and displacement MSCC indexes show completely different
Impact tests of the two damage structures were performed respec- robustness characteristic even their formats are same.
tively, and their macro-strain responses were processed for modal The third damage index, macro strain mode shape curvature
identification and subsequent damage detection. Three macro square change (MSCSC) as defined in Eq. (14), was also calculated
modal strain based damage indexes defined in Eqs. 12–14 were from macro strain modal shapes for damage detection of the D1
studied, and they were compared with corresponding indexes and D2 damage situations as shown in Fig. 19(b) and (d) respec-
using displacement mode shapes. tively. In contrast, Fig. 19(a) and (c) are MSCSC damage indexes cal-
The first damage index investigated is macro strain modal culated from displacement modal shapes for two damage
shape change (MSC) as defined in Eq. (12). Comparisons between situations respectively. This figure reveal the same finding that
undamaged state D0 and damaged states D1 in the 10% noise con- the macro-strain based index are much more robust to noise than
dition are plotted in Fig. 17(b) by using the strain mode shape of the displacement based one.
182 J. Zhang et al. / Engineering Structures 99 (2015) 173–183

-4
x 10
6 0.4
(a) 0% noise (b) 0% noise
5 0.5% noise 10% noise

Damage index (b)

Damage index (b)

0.3
4

3 0.2

2
0.1
1

0 0
1 10 20 30 40 50 60 66 1 10 20 30 40 50 6064
Girder1 Girder3 Girder1 Girder3
Node Number Element Number
-4
x 10
8 0.4
(c) (d)
Damage index (b)

Damage index (b)

6 0.3

4 0.2

2 0.1

0 0
1 10 20 30 40 50 60 66 1 10 20 30 40 50 6064
Girder1 Girder3 Girder1 Girder3
Node Number Element Number

Fig. 18. The MSCC index for damage detection, (a) displacement MSCC for D1 damage state, (b) strain MSCC for D1 damage state, (c) displacement MSCC for D2 damage state,
(d) strain MSCC for D2 damage state.

-5
x 10
1.2 0.1
(a) 0% noise 0% noise
(b)
1 0.5% noise 0.08 10% noise
Damage index (c)

Damage index (c)

0.8
0.06
0.6
0.04
0.4
0.02
0.2

0 0
1 10 20 30 40 50 60 66 1 10 20 30 40 50 6064
Girder1 Girder3 Girder1 Girder3
Node Number Element Number

-5
x 10
1.2 0.1
(c) (d)
1 0.08
Damage index (c)

Damage index (c)

0.8
0.06
0.6
0.04
0.4
0.02
0.2

0 0
1 10 20 30 40 50 60 66 1 10 20 30 40 50 6064
Girder1 Girder3 Girder1 Girder3
Node Number Element Number

Fig. 19. The MSCSC index for damage detection, (a) displacement MSCSC for D1 damage state, (b) strain MSCSC for D1 damage state, (c) displacement MSCSC for D2 damage
state, (d) strain MSCSC for D2 damage state.
 J. Zhang et al. / Engineering Structures 99 (2015) 173–183 183

6. Conclusions engineering, science, construction, and operations in challenging

environments. doi: http://dx.doi.org/10.1061/40988(323)164.
[4] Li ZX, Chan THT, Ko JM. Evaluation of typhoon induced fatigue damage for
Strain modal theory has been developed for a long time but it is Tsing Ma Bridge. Eng Struct 2002;24(8):1035–47.
far behind the development of the acceleration-based modal the- [5] Hoult NA, Fidler PRA, Wassell IJ, Hill PG, Middleton CR. Wireless structural
health monitoring at the Humber Bridge UK. Inst Civil Eng 2008;161(4):
ory. This is because it does not make sense to use the local strain
189–95.
measurement from point-type strain gauges for structural global [6] Brownjohn JMW, Pavic A. Experimental methods for estimating modal mass in
modal parameter identification. The developed long-gauge FBG footbridges using human-induced dynamic excitation. Eng Struct
2007;29(11):2833–43.
sensor is designed to measure the averaged strain over a long
[7] Hwang JS, Kim H, Kim J. Estimation of the modal mass of a structure with a
gauge length, and it can be connected in series to make a sensor tuned-mass damper using H-infinity optimal model reduction. Eng Struct
network for distributed strain measurement. Therefore, it has the 2006;28(1):34–42.
merit to measure not only local but also global structural informa- [8] Zhang J, Moon FL. Structural flexibility identification by integrating
substructures’ measurements. In: Proceedings of SPIE 2012: nondestructive
tion. This feature offers a new opportunity to develop the characterization for composite materials, aerospace engineering, civil
macro-strain modal identification theory. infrastructure, and homeland security, vol. 834708; 2012.
Based on the basic mapping relation between the long-gauge [9] Howell DA, Shenton III HW. System for in-service strain monitoring of ordinary
bridges. ASCE J Bridge Eng 2006;11:673–80.
strain and displacement, two methods have been proposed to iden- [10] Zhou YE. Fatigue problems in steel bridge structures. In: Proceedings of
tify macro strain modal parameters by processing the long-gauge structures 2004: building on the past, securing the future. doi: http://dx.doi.
dynamic strain. The first is based on the feature of the developed org/10.1061/40700(2004)44.
[11] Metzger AT, Huckelbridge AA. Temporal nature of fatigue damage in highway
strain FRFs, and the second is applying the singular value decom- bridges. ASCE J Bridge Eng 2009;14(6):444–51.
position to the strain FRFs. The example of a steel bridge model [12] Ye XW, Ni YQ, Wong KY, Ko JM. Statistical analysis of stress spectra for fatigue
investigated has successfully verified the effectiveness of them life assessment of steel bridges with structural health monitoring data. Eng
Struct 2012;45:166–76.
for macro strain modal identification.
[13] Zhou YE. Assessment of Bridge remaining fatigue life through field strain
Moreover, it has also been verified that the macro modal strain measurement. ASCE J Bridge Eng 2006;11:737–44.
based indexes are much more robust than traditional indexes using [14] Alampalli S, Lund R. Estimating fatigue life of bridge components using
measured strains. ASCE J Bridge Eng 2006;11:725–36.
displacement mode shapes. The damage detection results of the
[15] Shenton III HW, Hu X. Damage identification based on dead load
steel bridge model proved the usefulness of the long-gauge FBG redistribution: methodology. ASCE J Struct Eng 2006;132(8):1254–63.
sensor and related macro strain modal identification methods. [16] Necati Catbas F, Gokce Hasan B, Gul Mustafa. Nonparametric analysis of
Thus it is potential to apply them in engineering practices for structural health monitoring data for identification and localization of
changes: concept, lab, and real-life studies. Struct Health Monit 2012;11(5):
structural monitoring and safety management. 613–26.
[17] Xu ZD, Liu M, Wu ZS, Zeng X. Energy damage detection strategy based on
Acknowledgement strain responses for long-span bridge structures. ASCE J Bridge Eng
2011;16(5):644–52.
[18] Phares B, Lu P, Wipf T, Greimann L, Seo J. Field validation of a statistical-based
This work was sponsored by the National Science Foundation of bridge damage-detection algorithm. ASCE J Bridge Eng 2013;18(11):1227–38.
China (51108076). [19] Li SZ, Wu ZS. Development of distributed long-gauge fiber optical sensing
system for structural health monitoring. Struct Health Monit 2007;6(2):
133–43.
References [20] Wu ZS, Li SZ. Two-level damage detection strategy based on modal parameters
from distributed dynamic macro-strain measurements. J Intell Mater Syst
[1] Peeters B, Roeck GD. One-year monitoring of the Z24-Bridge: environmental Struct 2007;18(7):667–76.
effects versus damage events. Earthquake Eng Struct Dynam 2001;30(2): [21] Hillary B, Ewins DJ. The use of strain gauges in force determinations and
149–71. frequency response function measurement. In: Proceedings of 2nd IMAC;
[2] Melo M, Taveras J. Structural health monitoring of the golden gate bridge using 1984. p. 627–34.
wireless sensor network. Progress report, 10/08/2009. [22] Yam LH, Leung TP, Li DB, Xue KZ. Theoretical and experimental study of modal
[3] Katsuchi H, Yamada H, Kusuhara S. Structural monitoring and design strain analysis. J Sound Vib 1996;191(2):251–60.
verification of Akashi Kaikyo Bridge. In: Proceedings of earth & space 2008: