A Modified Wavelet Energy Rate Based

Damage Identification Method for Steel

Bridges
Mohammad Noori 1
Professor
International Institute for Urban Systems Engineering, Southeast University, Nanjing
210096, China, Mechanical Engineering and ASME Fellow, California Polytechnic State
University, San Luis Obispo, California, USA.
[email protected]
ASME Fellow

Haifegn Wang
PHD Student
State University of New York, Buffalo, NY 14260, USA
[email protected]

Wael A. Altabey
Assistant Professor
International Institute for Urban Systems Engineering, Southeast University, Nanjing
210096, China, Nanjing Zhixing Information Technology Company Nanjing, China,
Department of Mechanical Engineering, Faculty of Engineering, Alexandria University,
Alexandria 21544, Egypt.
[email protected]

Ahmad I. H. Silik
PHD Student
International Institute for Urban Systems Engineering, Southeast University, Nanjing
210096, China.
[email protected]

ABSTRACT

Strain is sensitive to damage, especially in steel structures. But traditional strain gauge does not fit bridge

damage identification because it only provides the strain information of the point where it is set up. While

traditional strain gauges suffer from its drawbacks, long-gage FBG strain sensor is capable of providing

1
Corresponding author. Tel.: +1-805-903-2411

1
the strain information of a certain range, which all the damage information within the sensing range can

be reflected by the strain information provided by FBG sensors. The wavelet transform is a new way to

analyze the signals, which is capable of providing multiple levels of details and approximations of the

signal. In this paper, a wavelet packet transform-based damage identification is proposed for the steel

bridge damage identifications numerically and with experimental experiment to validate the proposed

method. The strain data obtained via long-gage FBG strain sensors are transformed into a modified

wavelet packet energy rate index first to identify the location and severity of damage. The results of

numerical simulations show that the proposed damage index is a good candidate which is capable of

identifying both the location and severity of damage under noise effect.

Keywords: FBG strain sensor; Wavelet packet transform; Damage identification; Modified wavelet

packet energy rate.

INTRODUCTION

Because During the service life of steel bridges, they suffer from complex

working conditions, heavy working loads and other effects that will cause damages on

structures and accelerate the development of damages. With the accumulation of

damages on structures, the safety of both the structure and human life are threatened

and may directly result in tragedies of life and property [1, 2]. Therefore, structural

damage identification has become an important research topic in the engineering

protection field [3]. Over the last two decades the considerable development of

integrated monitoring systems for new and existing structures worldwide such as steel

bridges has been recognized as an important tool to solve structural damage

identification problem. It is commonly stated in the structural health monitoring

2
literature that damage identification approaches is classified as methods based on time

domain analysis [4], methods based on modal parameters [5] and methods based on

time-frequency domain analysis [6]. The majority of these techniques are based on

vibration which require data acquisition instruments to be fixed on the bridge directly,

which may be difficult and consuming of time, but it’s effective in giving a warning to

people if there is any indication of hazard on Bridge’s condition. These methods are

becoming a more important part of bridge monitoring systems.

In recent years, there are movement towards the development of indirect

vibration methods based on a vehicle response moving over a bridge. These methods

aimed to reduce the need of installing instruments directly on bridge to achieve more

efficient and low-cost. In related development Bu et al. [7, 8] investigated numerically a

bridge condition using the dynamic response of a vehicle running along a beam to

evaluate damage based on reduction of stiffness. The results showed that vehicle speed,

road surface roughness, measurement noise and numerical model errors have not a

significant effect on the accuracy of the method. Also González et al. [9] innovated a

novel algorithm using the vehicle acceleration as algorithm input to determine the

damping of a bridge. It was showed that the algorithm can be used for bridge stiffness

identification. It also was stated that the algorithm is not highly sensitive to signal with

low levels noise, the road roughness and model errors.

Furthermore, current damage detection techniques can be classified into two

broad categories: local and global damage identification. Most global damage

identification methods rely on damage introduced changes in the dynamic properties of

3
the identified structure .In these identification methods, system parameters such as

frequencies, deflected mode shapes, strain energy, flexibility matrix, etc. are used as

part of the damage index [10]. Among the discussed damage detection methods, some

are directly based on Fourier transform (e.g. natural frequency-based damage

identification methods). Fourier transform provides information in the frequency

domain and is not capable of detecting when (or where) a particular damage occurs. To

overcome this disadvantage, Dennis Gabor proposed the short-time Fourier transform

(STFT). This well-known windowing technique separates the signals into a series of

sections. Each time STFT is done within a small window that represents a section of the

signal. Thus, information on both frequency-domain and time-domain can be kept.

However, STFT has a disadvantage, the information about time and frequency (or space

and frequency) are acquired with a limited precision. According to the Heisenberg

uncertainty theory, this limitation is determined by the size of the time-frequency

window. Once the window size is set, it is the same for all frequencies, so a higher

resolution in time and frequency domain (or space and frequency domain) cannot be

acquired simultaneously.

The Wavelet Transform (WT) is a relatively new signal processing tool to analyze

data. Based on the theory of WT, it can be viewed as an extension of traditional STFT

transform with adjustable window location and size. With its unique merit of examining

signals with a "zoom lens having an adjustable focus", the original signal is separated

into two different levels of details and approximations [11-13]. Therefore, transient

information of the signal can be retained. The Wavelet Transform as technique for

4
structural damage identification, has been risen recently due to its capability of

capturing the transient behavior of a signal and analyzes it in time and frequency

domains and investigated by several researchers. Hester et al. [14] demonstrated the

ability of the wavelet transform to get information from time-frequency domains while

Zhao et al. [15-17] employed the structural mode shapes that extracted from the finite

element model of a simply supported reinforced concrete beam for damage

identification using different types of wavelets. Reda et al. [18] discussed in detail

various aspects related to wavelet transform as technique for structural damage.

Nguyen et al. [19] provided a method based on Symlet wavelet to evaluate bridge cracks

from vehicle displacement response. It was concluded cracks can be detected but higher

speeds give poor detection than low speed. Lee et al. [20] provided algorithm for truss

bridge based on the continuous relative wavelet entropy. . It was concluded that

damage can be detected but computation cost is very large for the real-life monitoring.

Although significant research has been carried out in the area of structural health

monitoring to make structures work safely, highly reliable and practical damage

identification methods are still lacking and many disasters occur due to lack of damage

identification.

Strain measurements have proved to be sensitive to damage. However, if

traditional strain gauges are not installed exactly on the damaged locations, the damage

detection result won’t be accurate. Thus, the idea of distributed sensing system was

proposed to overcome the limitation of the traditional strain gauge in obtaining the

strain, where they are installed [21]. Distributed sensing system is different from

5
multipoint sensing system. It is capable of capturing the information within a certain

range of the structure, leading to integrated information of the entire structure.

However, this requirement of distributed sensing systems was unreachable until

Horiguchi et al. [22] proposed the relationship between the strain and Brillouin

frequency shift in optical fibers. Based on this relationship, a concept of distributed

sensing system was proposed. Subsequently, Fiber Bragg grating (FBG) sensors [23],

Brillouin optical domain reflectometer (BOTDR) sensors [24, 25] and Brillouin optical

domain analysis (BOTDA) have been widely used in long-gage strain monitoring systems.

FBG sensors are based on the principle that the wavelength of the reflective

signal from the grating changes when it has longitudinal deformation. By measuring the

changes of wavelength, accurate deformation of the grating can be achieved. Based on

wavelength-division multiplexing technology, several FBG sensors can be combined in

one Fiber to achieve multipoint sensing. For Long-gage FBG sensors [26], fiber Bragg

gratings are packaged in series to extend the effective sensing length. FBG sensors have

their own merit: the signals used in FBG sensors are wavelength modulated signals,

which means that they don’t need to suffer from the limitations of inaccurate phased

measurement used in other kinds of fiber optic sensors.

BOTDR and BOTDA sensors are based on Brillouin scattering [24, 25]. Once there

is a change in strain or temperature on the fiber, the central frequency of Brillouin

scattering light changes. Based on the relationship, Brillouin scattering light can be used

to acquire changes in strain and temperature. For BOTDR and BOTDA sensors, any part

6
of the fiber is both a sensing unit and a signal transmission unit, leading to their

capability of spatial continuous measurement.

BOTDR/BOTDA sensors and FBG sensors have different merits based on their

sensing concept. FBG sensors have better accuracy in both dynamic and static sensing.

However, theoretically, FBG sensors are point sensors. Their sensing length is limited.

For BOTDR/BOTDA sensors, the accuracy and sampling rate is lower than FBG sensors.

But they are more suitable for distributed sensing, especially in large scale structures.

With the consideration of the small scale of the laboratory, long-gage FBG sensors are

adopted in this research.

In this paper, the goal is to investigate the effectiveness of a modified wavelet

energy rate based damage identification method for steel bridges elements (beam &

frame). Both simulated and experimental experiment were utilized to validate the

proposed method. Dynamic signals measured from structures are first decomposed into

the wavelet packet components. Then the modified wavelet packet strain energy rate

index is calculated based on the wavelet packet components, which is then used to

locate the damage and assess the severity of damage. Different scenarios are

considered to validate the proposed damage identification method. The simulated test

results show that the proposed damage identification method is able to detect both

location and severity of damages on a structure.

7
THEORETICAL BACKGROUND

Wavelet and continuous wavelet transform

Wavelet analysis is a signal processing method that relies on the introduction of

an appropriate basis and a characterization of the signal by the distribution of amplitude

on the basis [27]. If the wavelet is required to form a proper orthogonal basis, it has the

advantage that an arbitrary function can be uniquely decomposed and the

decomposition can be inverted. The wavelet is a smooth and quickly vanishing

oscillating function with good localization in both frequency and time. A wavelet family

 a,b t  is the set of elementary functions generated by dilations and translations of a

unique admissible mother wavelet  t  :

1 t b
 a ,b t     (1)
a  a 

Where b  R, a  0 are the scale and translation parameters, respectively, and t is time

(or location, if wavelet transform is utilized in spatial distributed signal). As the scale

parameter a increases, the wavelet becomes wider. So, each parameter shows the

signal at different scales and with variable time (or space localization).

The wavelet transform (in its continuous or discrete version) correlates the function

f t  with a,b t  . The continuous wavelet transform (CWT) is the sum over all time of

the signal multiplied by a scaled and shifted version of a mother wavelet:


t b
C a, b    f t  
1
dt (2)
a 
a 

8
 The results of the transform are wavelet coefficients, which determines how the

wavelet function signal expresses the signal. Hence, sharp transitions f t  create

wavelet coefficients with large amplitudes and this precisely is the basis of the proposed

damage identification method.

The CWT has an inverse: the inverse CWT helps to recover the signal from its

coefficients C a, b  and it is defined as:


t b 1
f t     C a, b
1
a ,b   2 dadt (3)
K a   a a

Where the constant K is:


ˆ  
2

K   d (4)
0  

Discrete wavelet transform (DWT) and Wavelet Packet Transform (WPT)

There is still a drawback of the CWT, which is a very large number of wavelet

coefficients C a, b  are generated during the analysis [28]. It can be shown that the CWT

is highly redundant, because it is not necessary to use the full domain of C a, b  to

reconstruct f t  . Therefore, instead of using a continuum of dilations and translations,

discrete values of the parameters are used in Discrete Wavelet Transform (DWT). The

dilation is defined as a  2 j and the translation parameter takes the values b  k 2 j ,

 j, k  Z and Z is a set of integers. It has been proved that DWT can fully decompose a

signal without losing any part of it. This sampling of the coordinates is referred to as

dyadic sampling because consecutive values of the discrete scales differ by a factor of 2

[29]. Using the discrete scales one can define the DWT:

9
 j  

 f  x   2 x  k  dx 

 f  x   x  dx
j
C j ,k  2 2
j ,k (5)
 

The signal resolution is defined as the inverse of the scale 1 a  2 j , and the

integer j is referred to as the level. As the decomposition level increase, the frequency

resolution increases and the smaller bandwidth components of the signal can be

obtained.

The signal can be reconstructed from the wavelet coefficients C j , k and the

reconstruction algorithm is called the inverse discrete wavelet transform (IDWT):

 
f  x   C j ,k 2 j /2  2 j x  k  (6)
j  k 

One possible drawback of the DWT is that the frequency resolution is quite poor

in the high-frequency region. It is difficult for DWT to discriminate signals containing

close high-frequency components. The wavelet packet transform (WPT) is one extension

of the DWT that provides complete level-by-level decomposition. The wavelet packets

are alternative bases formed by linear combinations of the usual wavelet functions. The

decomposed subcomponents can be treated as narrow-band signals under high

decomposition levels.

Wavelet packets consist of a set of linearly combined usual wavelet functions.

The wavelet packets inherit properties such as orthonormality and time-frequency

localization from their corresponding wavelet functions [30]. A wavelet packet is a

function with three indices,  ij, k t  where integers i, j , and k are the modulation, the

scale, and the translation parameter, respectively.

10
  ij ,k  t   2 j /2 i  2 j t  k  ,i  1, 2,. (7)

The wavelets  i are obtained from the following recursive relationships:


 2i  t   2  h  k  i  2t  k  (8)
k 


 2i 1
t   2  g  k  i  2t  k  (9)
k 

The first wavelet is the so-called mother wavelet function:

 1 t    t  (10)

The discrete filters h(k ) and g (k ) are quadrature mirror filters associated with

the scaling function and the mother wavelet function, which can be treated as low-pass

and high-pass filter. A quadrature mirror filter is a filter whose magnitude response is

the mirror image around  / 2 of that of another filter. There are quite a few mother

wavelets reported in the literature. Most of these mother wavelets are developed to

satisfy some important property such as the invertibility and the orthogonality.

Daubechies developed a family of mother wavelets based on the solution of a dilation

equation. One of these wavelets, DB20, is adopted in this study.

Similar to the FT, any measurable and square-integrable function can be

decomposed into wavelet packets. The decomposition process is a recursive filter-

decimation operation. Fig. 1 shows a full WPT tree of a time-domain signal f t  up to

the 3rd level of decomposition. Fig. 1 shows that the DWT consists of one high-frequency

term from each level and one low-frequency residual from the last level of

decomposition. The WPT, on the other hand, contains complete decomposition at every

11
level and hence can achieve a higher resolution in the high-frequency region. The

recursive relations between the j th and the j  1th level components are:

f ji (t )  f((2j i1)1) (t )  f(2ji1) (t ) (11)

f j2i11  t   Hf ji  t  (12)

f j2i1  t   Gf ji  t  (13)

Where: H and G are filtering-decimation operators and are related to the discrete

filters hk  and g k  through:


H t    hk  2t  (14)
k  


Gt    g k  2t  (15)
k  

After j levels of decomposition, the original signal f t  can be expressed as:

2j
f  t    f ji  t  (16)
i 1

The wavelet packet component signal f ji t  can be expressed by a linear

combination of wavelet packet functions  ij, k t  as follows:


f t  
j
i
c
k 
i
 ij ,k  t 
j ,k (17)

The wavelet packet coefficients cij , k can be obtained from:


cij ,k   f  t   t  dt
i
j ,k (18)


Providing that the wavelet packet functions are orthogonal:

12
  mj ,k  t  nj ,k  t   0, if m  n (19)

Each component in the WPT tree can be viewed as the output of a filter tuned to

a particular basis function, thus the whole tree can be regarded as a filter bank. At the

top of the WPT tree (lower level), the WPT yields good resolution in the time domain

but poor resolution in the frequency domain. At the bottom of the WPT tree (higher

level), the WPT results in good resolution in the frequency domain yet poor resolution in

the time domain.

Modified Wavelet packet energy rate

The continuous wavelet transform (CWT) Wavelet Packet Energy

Energy concentration is one of a reliable feature for time–frequency data

processing. Particularly, energy has been used successfully for classification application.

Thus, the wavelet packet energy representation can provide more robust signal features

for classification, while it is difficult to identify these features directly from the

expansion coefficients [31, 32]. Furthermore, wavelet packet energy can be used to

identify the locations and severity of damage. To do that, the wavelet packet energy

E n j at j level of node n is defined as:

 2j 2 j  2j
En j   f  t  dt    f t  f t  dt  E
2 m n
j j nij
(20)
 m 1 n 1  i 1

A signal f t  should be square integrable if En is finite, that is, f t  is in the

Hilbert space L2 R  .

13
Note that: energy in this context is not the same as the conventional notion of energy in

other areas of science .The signal energy in the j th level and the i th frequency band can

be expressed as:

n 2

E j   C j k  , j  1,2, , m (21)

k 1

The total energy of signal is expressed by:

2j
Ef j   Ef i (22)
j
i 1

Where: the wavelet packet component energy E ni can be considered to be the energy
j

i
stored in the component signal f j (t ) :


Eni   f  t 
i 2
j dt (23)
j


Eq. (20) demonstrates that, the overall signal energy can be spilt to wavelet

packet energy components in various, frequency bands. Finally, the WPERI was

developed as follows, to identify the location and extent of the damage of the crack

[33]:

E   E 
 
2j

E 
f ji b f ji a
 Ef j   (24)
i 1 f i
j a

  is the signal component energy at level j without damage, E 

The term E f i
j a f ji b

is the signal component energy at j level with some damage. It is assumed that

structural damage would influence the wavelet packet component energies and then

may change this damage indicator. Therefore, it’s desired to choose the Wavelet Packet

14
Energy Rate Index (WPERI) because it is sensitive to the alerts in the signal

characteristics [34].

Modified Wavelet Packet Energy

Structural damage detection utilizing Wavelet Packet Energy is one of the most

reliable and efficient structural health monitoring techniques. The Energy methods

which are based on strain response data is rather sensitive than one based on

acceleration, displacement, because strain. Energy density involves the second spatial

derivatives of the displacement, so it is much more sensitive to small deviation in the

structural response than the displacement itself. Therefore, Wavelet Packet Energy has

been modified to include the effect of strain in order to increase the accuracy of

damage detection. It is expected that deploying the modified method can obtain more

accurate strain energy which is stored in structural elements and at the end provides

appropriate, damage detection model and reduces the computation and iteration

efforts.

Strain is more sensitive to damage than other raw data, such as deflection,

velocity and acceleration. In certain cases, damage can be directly identified via raw

strain data. For structural dynamic responses, the envelope area of strain-time

curvature (EASC) is proposed as a damage index:


Sn  f

n (t )dt (25)

15
 To quantify the damage from wavelet packet component energies, the modified

wavelet packet energy rate (MWPER) is proposed in this paper to detect both the

location and the severity of the damage. MWPER of node n is defined as:

    E     S    S  

2j
Dn     Eni (26)
i 
 j a
nij
b
n a n b

Where subscripts a and b stand for damaged, and undamaged status, respectively.

Wavelet packet energy rate is capable of utilizing information extracted from

different frequency-bands, and the envelope area of strain-time curvature extracts the

amplitude information from strain time-history data. Thus, MWPER can utilize both the

frequency domain information and amplitude information of the raw data.

A procedure of damage identification based on the modified method is assumed

that the reliable intact and damaged structural models are available and the structure is

excited via the same impulse load and acts at the same location in damage and

undamaged structure.

DAMAGE IDENTIFICATION PROCEDURE

Based on the discussion above, a complete damage identification procedure is

established for the second method. Fig. 2 shows the complete procedure of damage

identification based on WT. The steps for this procedure are described below.

Establishment of numerical model

Developing a numerical model is the first step for the detection of the severity of

the damage. Thus, a fine mesh numerical model of the structure should be established.

Simulation of running car experiments. At least four points of damage needs to be

simulated.

16
a. Damage at point DB1 with a stiffness loss of SL1

b. Damage at point DB1 with a stiffness loss of SL2

c. Damage at point DB2 with a stiffness loss of SL1

d. Damage at point DB3 with a stiffness loss of SL1

Based on the MWPER data from simulations a and b above, the near-linear

relationship between the stiffness loss and MWPER can be determined. Based on the

MPWER data of a, c and d, the relationship between the location and MPWER at a

certain level of damage severity can be determined.

Establishment of the damage index map

Based on the relationships obtained in the previous section, a damage index map

is established.

Denoising of the input signal

The original data used in this research is long-gauge strain data obtained via

FBG/BOTDR sensors. As discussed before, denoising is commonly the first step for signal

processing. Then all input signals will be denoised by WT.

Calculation of MWPER

Based on the value of MPWER, the damage location can be found.

Combine the damage location and the value of MWPER

Now that a map where all the relationships have been established is developed,

the severity of the damage can be determined by examining the corresponding location

and the data from MPWER.

Maintenance decision based on the damage location information

17
 Once the damage location and its severity are verified, it can be determined if

the structure needs to be repaired.

VALIDATION BY NUMERICAL MODELS

A completed theory for damage identification scheme needs to be validated by

experiments. Thus, numerical simulations are essential for the development of the

proposed method. The theoretical background has been elaborated in the previous

section, hence, numerical simulations are introduced herein to validate this method.

Numerical model

In this section, MWPER is introduced as a damage index, whose calculation is

based on WPER and EASC and the calculation of wavelet packet energy is based on

strain time history data of the structure. Since the entire damage identification is based

on the strain time history data, the model used in this section should be able to have

enough strain at every part of the structure. Although a simply supported beam won’t

have any strain at the support points, the supporting condition of fixed ends and simply

supported are both considered and compared to see if the proposed method can

identify damages in the vicinity of the supports of a simply supported beam. In this

work, both beam and single layer frame models will be used to valid the proposed

damage identification.

ANSYS 14.5 is introduced as the numerical simulation platform. For the

convenience of modeling, all the material properties and cross sections are set to be the

same among all models. The material properties used herein have been commonly used

18
in numerous research work, including: E  200GPa ,   7800kg / m3 and   0.3 , where

E stands for Yong’s modulus,  stands for mass density and  stands for Poisson’s

ratio.

All three numerical models are described in detail below. For the convenience of

comparison and modeling, undamaged and damaged sections of beams and columns in

all 3 models are set to be the same. A total of 8 section scenarios are considered in this

paper for the beam model. The beam section is set to be a thin-walled rectangular tube,

whose height, width and thickness are 70mm, 50mm and 4mm, respectively (Fig 3 (a)).

The column section is set to be a square steel tube, whose height, width and thickness

are 50mm, 50mm and 4mm, respectively (Fig 3 (b, c)). Corrosion or crack on bridges

both reduce the working area of the component sections, thus we simulate the damage

via reducing wall thickness. Since hydrops cause corrosion at the bottom wall of a

structure, the thickness of bottom wall will be reduced to simulate damages. The

reduced part of the bottom walls is shown in Fig 3 (a) and Fig 3 (b, c) as the shaded

parts.

All damages/cracks are simulated via reducing the thickness of the bottom layer.

The stiffness loss should be calculated exactly, since the wavelet coefficients might be

used as a damage index that are able to reflect the damage severity. For the purpose of

simulating the damages with different damage level, a formula is proposed to calculate

the exact thickness loss that is needed for a certain bending rigidity loss. The second

moment of area is:

19
 Tt 3W Tt 2 (Tb  d )3W T d 2
I  TW
t (H n  )   W (Tb  d )( H  H n  b ) 
12 2 2 2
Tl ( H  Tt  Tb  d )3  ( H  Tt  Tb  d )
2

 Tl ( H  Tt  Tb  d )   Tt  H n   (27)
12  2 
Tr ( H  Tt  Tb  d )3  ( H  Tt  Tb  d )
2

 Tr ( H  Tt  Tb  d )   Tt  H n 
12  2 

H n and all the parameters, such as W , H , d , Tt , Tb , Tl and Tr are shown in Fig. 3

(b, c) ,

where: H n  WH  H  d H  d  Tt  Tb W  Tl  Tr  WH  W  Tl  Tr H  d  Tt  Tb 

 2

 2 2 

With this formula, a damage range from zero to 30% bending rigidity loss is

proposed for the following sections. Table 1 shows the exact thickness loss of the

bottom layer.

All beam section information are listed in Table 1 For column sections, only one

damage severity level is used, the stiffness loss of that damage is 3.81%.

D in the section plots (Fig. 3 (b, c)) stand for the damage depths, and for the

convenience of description, damage depth zero means there is no damage at the

corresponding section (Section ID 1).

The damage location information for the adopted numerical models are

introduced the following detail.

Model 1: Simply Supported beam

The simply supported beam’s length is set to be 1000mm, with a damage

introduced at distances of DB from the left support as shown in Fig. 4 (a).

20
Model 2: Fixed ends beam

The fixed ends beam’s length is set to be 1000mm, which is the same as the

simply supported beam. The damage location parameter DB is defined as the distance

between the damage location and the left support as shown in Fig. 4(b).

Model 3 Single Layer Frame

The single layer frame shown in Fig. 4(c) is used as the 3rd numerical model.

Damages are marked by rectangular symbols in Fig. 4(c) DB is the distance of the

damage location from the centroidal axis of the left column. LB is the total length of the

beam. DCL is the distance from the centroidal axis of the left column, while LCL is the

total length of the column. The unit used in Fig. 4(c) is mm.

Damage Scenarios

For the establishment of the second method, several different damage scenarios

need to be taken into consideration. There are three cases for the establishment and

validation for that method.

Firstly, two different kinds of excitations are compared, which is simulated via

contact elements in ANSYS. So far, most dynamic experiments are done by hammer

excitation. For bridge, cars run on the top of the beam. Both running car excitations and

hammer excitations are usually used for dynamic experiments on bridges, thus these

two types of excitations are adopted in this work. The hammer is simulated by a mass of

8kg, and it hits the structure at a speed of 4m/s. The running car is simulated by a mass

of 8kg and the car runs from the left side to the right side at a speed of 2m/s.

21
 Secondly, the situation that damages occur at points where the strain is

relatively small must taken into consideration to validate if the proposed method can be

used independently without additional detection method that can detect. As mentioned

in previous sections, the proposed method only works in the context that strain exists in

every part of the structure. For steel bridges, although stain exists everywhere on the

structure, there are always some points where the strain is relatively small, which makes

it essential to validate the proposed method for small strain situation.

Thirdly, a multiple damage scenario must be taken into consideration, since the

damages/cracks on steel bridges may happen simultaneously. Although the theory

behind the second damage identification method indicates that the spatial resolution of

the identification is limited by the density of sensors and the sensing range of each

sensor, the multiple damage scenario must be validated to make sure the method works

well with steel bridges.

Finally, the single layer frame with damages on columns must be tested, since

the excitation range is limited within the main beam. While columns are not directly

excited by excitations, this damage scenario must be considered and validated.

Based on the discussion above, four damage scenarios are considered and used

for the establishment and validation of the second method. For the convenient of

analysis, the sensor interval is set to be the same as element size, and the default

element size is set to be 12.5mm. To identify the effect of sensor interval, there are

three element sizes in scenario 2, 12.5mm,50mm and 200mm, which means there are

80, 20 and 5 sensors on the beam of scenario one, respectively.

22
1. Hammer excitation on a fixed end beam with one damage at position DB=0.25m;

2. Running car excitation on a fixed end beam with one damage at position DB=0.25m,

and the car runs from the left support to the right support at a speed of 2 m/s;

3. Running car excitation on a simply supported beam with one damage at position

DB=0.025m and DB=0m. The car runs from the left support to the right support at a

speed of 2 m/s. Since numerical models consist of many elements, the so called

DB=0m is simulated by the section change of the element on the left side of the

beam;

4. Running car excitation with 2 damages on the main beam of a single layer frame

model and one damage on the left front column, the damages occur at DB1=0.5m,

DB2=0.1m and DCL=0.5m. The speed of the running car is 4m/s, and the car runs

from the left end of the main beam to the right end of the main beam.

For all scenarios, the strain data of different sensors are recorded

simultaneously. The record for scenario 1 begins at 0.5 before the excitation and lasts

for 5 seconds. The record time for scenario 2 and 3 is 4.5 seconds, and the record starts

when the car starts to move from the left support. For scenario 4, the record starts

when the car starts to move from the left support, and also lasts for 4.5 second.

The establishment of the proposed method

Based on the damage identification theory proposed previously, several issues

must be taken into consideration and modified to establish a reliable damage

identification scheme. Based on this idea, the following parts are analyzed to refine the

damage identification theory into a complete damage identification scheme.

23
 In this section damage scenario 1 is used to present the effectiveness of MWPER

by comparing modified wavelet packet energy rate (MWPER), wavelet packet energy

rate (WPER) and Envelope Area of Strain-time Curvature (EASC).

Comparison of the proposed damage index and conventional indices

A single damage with 3.52% stiffness loss is considered and the damage is

assumed to be at the 20th and 21th elements which are located 0.25m away from the left

support. All 3 damage indices are normalized to 1 for the purpose of making the peak

values the same (Fig. 5).

All 3 damage indices are capable of indicating the existence of damage in

scenario 1. While values of WPER and EASC are much larger than MWPERR on elements

are intact, especially on elements near the supports, which indicates that WPER and

EASC are not so stable when applied on small damage cases and may lead to a false

indication of damage in some cases where damage severity is low. Compared with

WPER and EASC, MWPER is stable with a disturbance less than 0.05 on intact elements.

Comparison between hammer excitation and running car excitation

In this section scenario 1 and 2 are used to choose the best excitation mode for

the proposed damage identification. All damages are set to be at DB=0.25, and the

stiffness loss is set to be 3.52%. For hammer excitation, the hammer hits the beam at a

speed of 4 m/s at time 0.575s, and the recording lasts for 4.5 seconds. Fig. 6 shows the

strain time history data of the midpoint. Fig. 7 shows the instantaneous strain data of

the entire beam at 0.01s after the excitation by the hammer.

24
 For running car excitation, the recording starts when the car starts to move from

the left support and ends at 3.925 seconds after the car leaves the right support. The

recording lasts for 4.5 seconds, the same as the hammer excitation. Fig. 8 shows the

strain time history of the midpoint of the beam. Fig.9 shows the instantaneous strain of

the fixed end beam at 0.01s after the car runs off the main beam.

Fig. 10(a) shows the comparison between hammer excitation and running car

excitation. Results show that the running car result is much better than the hammer

excitation, especially at the supports. While the damage index of the hammer excitation

reaches 0.05 at the supports, that of the running car excitation is almost zero, which

means that the MWPER of a running car excitation is more stable than that of a hammer

excitation.

The abovementioned phenomenon is caused by the number of excitation points.

The proposed method is capable of combining all the strain time history data into one

index for damage identification, which means the more points that has large strain, the

more stable result we can obtain. With a running car excitation, every part could be

excited, resulting in a large number of points that has large strain, which means better

damage identification. Thus, for damage identifications in a steel bridge, it is the best

that cars run from the left end to the right end, i.e., the entire span of the beam, in

order to collect enough response data.

Effects of different sensor intervals

In this section scenario 2 is used to examine the effect of sensor interval on the

proposed damage identification. All damages are set to be at DB=0.25, and the stiffness

25
loss is set to be 3.52%. There are three sensor intervals, 12.5mm, 50mm and 200mm,

which means each sensor covers 12.5mm, 50mm and 200mm length of the beam,

respectively.

As Fig. 10(b) shows, the proposed damage index is capable of identifying damage

with five sensors on the beam, where the sensor interval is 200mm. The identified

damage locations are different. With 200mm sensor intervals, the identified damage

location is about 300mm from the left support; with 50 mm sensor intervals, the

identified damage location is about 225mm from the left support; with 12.5 mm sensor

intervals, the identified damage location is about 250mm from the left support. While

the actual damage location is DB=250, the damage location identification errors can be

explained and are acceptable. This phenomenon is caused by the sensors’ mechanism.

For long-gage fiber optical sensors, the strain obtained are average strain of the whole

measure length, which means as long as the damage location covered by sensors

applied, the strain change can be captured by the corresponding sensor. Since the x

coordinate for each data point in Fig. 10(b) is set to be the mid-point of corresponding

sensor coverage length, the peak in Fig. 10(b) means there is a damage or damages in

the corresponding sensor interval. Under the context that only 5 sensors or 20 sensors

set up on the beam, Fig. 10(b) shows that the proposed method is capable of identifying

damages with only a few sensors.

Detectability of damages nearby the supports of the simply

In this section damage scenario 3 is used to test if the second damage

identification method is able to detect damages near the supports of a simply supported

26
beam. The second damage identification method is based on strain signals. Since the

strain values at the points near the supports of a simply supported beam are much

lower compared with those at the points in the middle of the span, it becomes the main

concern if damages occurring at points where the strain signals are weak can be

identified.

A stiffness loss of 3.52%, the same as in the previous sections, is introduced on

the simply supported beam. The car is run from the left side to the right side at a speed

of 2m/s. One damage is set on the left end of the simply supported beam. Another

damage is set to be at a distance of 0.025m from the left support. Fig. 11(a) shows that

the damage at DB=0.025m can be identified, while the damage at DB=0m cannot be

identified.

All the MWPER are normalized in order to get the best estimation of the damage

location. Thus, the absolute values in Fig. 11(a) are not our concern, only the relative

values need to be of concern. The reason behind the fact the strain in the elements near

the supports of a simply supported beam are not totally zero, while strain of the

elements at the support is almost zero seem to be that all data in Fig. 11(a) are

normalized. With FBG/BOTDR strain sensors, the result that are obtained represent the

information gathered from the entire area covered by the sensors. Therefore, the strain

change, caused by the damage, at DB=0.025 is detectible, while that of the damage at

DB=0m is covered by other parts, whose strain value is much larger.

Since damages close to the hinge support can be identified, a conclusion can be

drawn that as long as the damaged area is covered by a long-gauge strain sensor, and

27
the strains on that area are not all zero, damages that located at small strain areas can

be identified. However, if the strain is too small, such as at the elements located at the

supports of a simply supported beam, the damage will not be detected.

Detectability of multiple damages on the main beam and the damage on the column

As the next step, we explore if the concept behind the second method can be

validated by applying this approach for multiple damage situations. In this section this

proposed concept is validated by the single layer frame model with two damages on the

main beam.

Since the running car excitation can only be applied on the main beam of the

single layer frame, which means the columns cannot be excited directly, we will explore

how damages on the columns can be taken into account and detected. Damage scenario

4 is used to test if the proposed method can be used for this case.

Fig. 11(b) and Fig. 11(c) show that all three damages are identified. Fig. 11(b)

indicates that multiple conditions are applicable for MWPER, which is predictable since

the strain is sensitive to local damage. Fig. 11(c) shows that damages on a component

that is not directly excited, i.e. the left column, is also detectable for MWPER. Thus,

MWPER is an effective detection tool for application in steel bridges since all damages

on all components can be identified.

Verification of the noise effect on the proposed method

All the results presented in this section are based on strain time-history data

obtained from the finite element analysis of the response, and hence contain no

experimental noise. For real cases, experimental noise is inevitable. To evaluate the

28
robustness of MWPER under measurement noise, the simulated data of Scenario No.4

are contaminated with certain level of artificial random noise to generate ‘measured’

data. Normally, distributed random noises whose amplitudes are 5%, 10%, 30% and 50%

of the root-mean-square (RMS) value of strain data, respectively, are added to the strain

time-history data.

MWPERs in Figs. 12 (a-d) are normalized to make the comparison clear. With the

increase of the noise level, disturbance on intact elements increase, especially elements

near the end points of the beam. It is important to note that even under a noise level of

30% the damage can still be identified. This demonstrates that the second damage

identification method is robust enough to take into account the measurement noise.

Effects of stiffness loss level

As stated in the previous sections, MWPER is capable of identifying the location

of a damage under a certain noise level, which means the 2nd level of damage

identification. In order to accomplish the 3rd level of damage identification, and to show

the ability of MWPER to quantify the damage, in this section MWPERs of different

stiffness loss levels (3.52%,5%,10%,15%,20%,25%,30%) are discussed.

Fig. 13 shows the MWPERs on the 20th element changes with the change of

stiffness loss. As can be seen, the MWPERs also show a near-linear relationship with the

stiffness loss. Thus, with more stiffness loss level of finite element data, MWPER is

capable of identifying stiffness loss level of a specific position. For a damage

identification method that can detect both the location and severity of a damage, the

relationship between the damage index and the location needs to be made clear. As

29
demonstrated in this case, the relationship between the location and the value of

MWPER is estimated. Therefore, this approach is capable of detecting the severity of the

damage as well.

Effects of damage location

Quantification of the damage needs a precise numerical relationship between

the damage severity and the damage index. In the previous section, it was verified that

MWPER has a near-linear relationship with stiffness loss. Therefore, the effects of the

damage location need to be taken into consideration. In this section, damage scenario 2

is considered in order to explore and discuss the effects of damage location with the

damage simulated by a stiffness loss of 3.52%.

Based on the idea that both the damage location and the damage level affect

MWPER, the effects of damage location must be figured out in order to make this

damage identification method capable of determining damage level. Subsequently, a

damage spanned across the location DB=0 to DB=1 are simulated to determine the

exact effects that the damage location will have on MWPER. Eighty damage locations

were simulated to obtain the acquired results. Fig. 14(a) shows the MWPER of damaged

beams with damage locations spanning across DB=0 to DB=1.

Consideration of both damage location and damage level

In this section both a simply supported beam and a fixed end beam are used for

the calculation the relationship map among the location, damage severity and location.

Eighty damage location situations and 6 levels of stiffness loss are considered, which

means that each beam’s numerical model needs to be simulated for 80*6=240 times. In

30
addition, there are two support conditions leading to 240*2=480 times of simulation.

This requires an exhaustive computational time. Thus, for utilizing ANSYS for this

computation, a computation platform is set up which consists of an Intel 4790K, 16GB

DDR3 2200MHz and a RAID0 disk array, in order to carry out a fast computation. Finally,

these simulations form two damage index maps, which can be used for the

identification of damage severity.

As can be seen in Figs. 14(b, c), the support conditions have an effect on the

distribution of MWPER. Thus, for different structures different MWPER maps need to be

established for the purpose of damage severity identification.

Consideration modeling error

As discussed in the previous section, the identification of damage level needs a

MPWER map generated by FEM simulation. The modeling-error must be taken into

consideration, especially for its effect on the accuracy of damage severity identification.

Based on scenario 2, 6 modeling errors are tested. The stiffness of the simulated beam is

changed to 80%, 90%, 110%, 120%, 130%, 150% of the original stiffness and tested. The

result MWPERs are showed in Fig 15. The MWPER has a nearly-linear relationship with

the stiffness change, almost the same as the index-damage relationship. The similar

nearly-linear relationships can be explained by the mechanism of the proposed method.

Both damage or modeling error will lead to the change of strain signal. And their effects

on strain amplitude are almost the same as each other, which means both the strain

change caused by damage and the strain change caused by modeling error will lead to

the increase MWPER.

31
 Fig. 15 and the discussion above show that modeling error has a large impact on

the accuracy of the proposed method. The modeling error must be limited with a small

interval to ensure that the damage severity identification result is accurate.

Validation by scaled single layer frame

In this section, scenario 4 is used for the validation of the proposed damage

identification scheme. According to the result of section 4.3.3, location of all damages

can be identified. Thus, the only information that needs to be obtained and validated is

the capability of this approach in identifying the damage severity.

MWPER values of two damages on the main beam will be used as the reference

to establish a damage index map. Subsequently, the index map will be utilized to

identify the severity of the damage on the left front column.

Fig. 16 shows the damage index map established for the left front column. As

can be seen, its shape is between a simply supported beam and a fixed ends beam.

Based on the damage index map, the severity of the damage on the column can be

identified. As noted earlier, the MPWER of the damage point on the column is 1.836E-4

and the damage location is at DLC=0.5m (Fig. 15). Based on these two coordinates, the

third coordinate in Fig. 16 can be found, which is 22.36%. While the stiffness loss in

simulation is set to be 21.35%, the stiffness loss value obtained via the MWPER map is

22.36%. Thus, the damage severity can be identified by MPWER.

32
VALIDATION BY EXPERIMENT

The best way to determine if a damage identification method is good is always

by testing it on a real structure. Thus, in this work an experiment is designed and run to

determine whether the damage identification method works well. The experimental

result showed that the proposed method is capable of identifying the damage in a real

structure.

Design of the experiment

In order to validate the effectiveness of the proposed damage identification

method, a simply supported steel beam is set up. The steel beam is an H-shape steel

beam. The section of the bean is shown in Fig. 17, where the length unit is mm. The

length between two supports is 6000mm (Fig. 18(a)).

Usually the fixed ends supporting condition is difficult to reach, limited by the

fixation method. Besides simply supported support condition will lead to lower strain

values nearby supports, which is helpful for testing the efficiency of the proposed

method. Hence, a simply supported beam is set up as the main structure of this

experiment.

As discussed earlier, commonly used long-gage strain sensors are FBG sensors

and BOTDER/BOTDA sensors. Both of them are capable of acquiring long-gage strain

information from the structure. FBG sensors have better accuracy in both dynamic and

static sensing, while FBG sensors are limited by the sensing range. BOTDR/BOTDA

sensors are suitable for long range sensing, such as a range of over several kilometers.

33
But the accuracy of BOTDR/BOTDA sensors is lower than FBG sensors. Considering that

the steel beam is only 6 meters long, FBG sensors are used in this experiment.

The sampling rate is usually determined by the natural frequencies of the tested

structure and sensing system properties. FBG sensors’ sampling rage can reach an

extremely high value, thus, it’s not the limitation for the decision of sampling rate. And

then the natural frequencies of the tested structure should be considered as the only

aspect that determines the sampling rage. According to numerical experiments, the first

three natural frequencies are 6.51 Hz, 10.65 Hz and 31.94 Hz. Since the first three

natural frequencies are the most important for analysis, the sampling rate is set to be

1000Hz, which is high enough to capture all vibration features of the beam.

This experiment is aimed to obtain sufficient data. Hence, sensors, such as

accelerometers and displacement meters and FBG strain sensors are installed on the

tested beam. For the convenient of installation, FBG sensors are pasted at the bottom

surface with displacement meters, while accelerometers are installed on the top surface

of the tested beam. As can be seen in Fig. 18 (a), many sensors are distributed on the

top surface and bottom surface of the tested beam. Thus, running car excitation is

impossible for this tested beam. As a result, point excitation is adopted in this

experiment. The excitation is introduced by a force hammer, which helps record the

hammer force for the experiment.

In this experiment, the damage on the beam is introduced by cutting holes at top

flange (Fig. 18 (b)). Since these holes cannot be cut accurately, the stiffness loss of the

34
damage section is calculated after the cutting procedure. By accurate measurement and

calculation, the stiffness loss of the damaged section is determined to be 6.6%.

Experiment process

The FBG sensors used in this experiment have a sensing range of 1.5m. Based on

sensor parameters, number of sensors, size of the tested beam, the tested beam is

separated into 12 segments (E1, E2… E12), each has a length of 0.5m (Fig. 19). The

dividing points are used as excitation points, thus, there are 13 excitation points (H1,

H2… H13). All these excitation points are shown in Fig. 19. The damage is introduced on

E3 by cutting holes on its top surface (Fig. 18 (b)).

The test procedure run 24 times. As listed in Table 2, four excitation points were

selected. Six tests were recorded for each excitation point, three for the intact status

and three for damaged status.

In order to match the damage index map, all strain signals are normalized by

setting the hammer force to 1. Table 2 shows the hammer force record of all these

tests. Recorded data of test 5 was shown in Figs. 20-21. Fig. 20 shows the hammer force

time history of test 5, which quite clear. As can be seen in Fig. 21 the strain signal is

contaminated badly, with an error of about 10% of the largest strain. Fig.22 shows the

instantaneous strain of both damaged and intact beam, and there is no signature that

stands for the damage. Thus, the proposed damage identification method is utilized.

35
Damage identification based on the data acquired by the experimental steel beam.

In this section the data acquired during these experiments are applied to the

proposed damage identification scheme. Fig. 22 indicates that the instantaneous strain

distributions cannot show the damage location, while instantaneous strain distribution

shows the damage location by slight disturbance on the curvature (Fig. 7 and Fig. 9). This

phenomenon is caused be noise effects, which is shown clearly in Fig. 21.

A damage index map is presented for simply supported beam (Fig. 14(a)).

However, the excitation conditions are different, thus, a new damage index map is

calculated and presented in Fig. 23.

As can be seen, all experiments’ data can be utilized to identify damage locations

(Fig. 24). The damage is shown clearly by peaks on element 3. But the MPWER value of

each damage identification is different form another. This situation is callused by noise

effect since the noise cannot be denoised completely.

The next step of damage identification is identifying the damage severity. As is

known that the damage locates on element 3, namely location 1.5m. Although the

MWPER values of different tests are different, they are considered to have the same

importance to the damage identification result. Thus, the mean value of these 12

MWPER values is used for severity identification, which is 0.0407. Now two coordinates

of the damaged point on the damage index map are determined. Fig. 25 shows the

MWPER function at position 1.5m. As can be seen in Fig. 25, the stiffness loss level is

10%, while the accurate damage level is 6.6%. Although there is a large difference

36
between the identified stiffness loss and the accurate damage level, considering the

noise effects in measured strain signals is relatively high, this result is acceptable.

CONCLUDING REMARKS

In this paper, MWPER has been proposed for steel bridge damage identification

based on long-gauge fiber optic strain sensing. With both numerical and experimental

validation, the proposed method proved to be suitable for the health monitoring of the

entire structure. Moreover, the proposed method is robust to noise. As presented, the

damage location can be identified under a noise level of up to 30%. Furthermore, it was

clearly shown that the proposed scheme can be applied to identify multiple damage

locations. It was also illustrated that the running car excitation is better than hammer

excitation for the proposed damage identification method.

Based on the near-linear relationship between MWPER and the stiffness loss and

the relationship between MWPER and the location, three damage index maps for

damage identification are established for simply supported beam, fixed end beam and

the left front column of the single layer frame. By utilizing these two maps, a

corresponding map for single layer frame damage identification is established and

tested. The test results show that the proposed damage identification method is able to

detect both location and severity of damages on steel bridges.

ACKNOWLEDGEMENT

This research was supported by IIUSE of Southeast University, by a grant

provided by 1000 Program for the Recruitment of Global Experts. These support are

gratefully acknowledged.

37
REFERENCES

[1] Padgett, J. E., and Tapia, C., 2013, "Sustainability of natural hazard risk mitigation:
Life cycle analysis of environmental indicators for bridge infrastructure", J. Infrastructure
Systems, 19(4), pp.395-408. https://doi.org/10.1061/(ASCE)IS.1943-555X.0000138.

[2] He, C., Xing, J., Li, J., Qian, W. and X. Zhang, 2015, "A New Structural Damage
Identification Method Based on Wavelet Packet Energy Entropy of Impulse Response",
the Open Civil Engineering Journal, 9, pp. 570-576.

[3] Balageas, D., Fritzen, C. P., and Güemes, A., 2006, "Structural Health Monitoring",
Wiley-ISTE, London, pp.493. ISBN: 978-1-905209-01-9.

[4] Jiang, S. F., Wu, S. Y., and Dong, L. Q., 2014, "A time-domain structural damage
detection method based on improved multiparticle swarm coevolution optimization
algorithm", J. Mathematical Problems in Engineering, 20(14), Article ID 232763.
http://dx.doi.org/10.1155/2014/232763

[5] Shi, Z. Y., Law, S. S., and Zhang, L. M., 2000, "Structural damage detection from
modal strain energy change", J. Engineering Mechanics, 126(12), pp.1216-1223.
https://doi.org/10.1061/(ASCE)0733-9399(2000)126:12(1216).

[6] Liu, Y., Fard, M. Y., and Chattopadhyay, A., 2012, "Damage assessment of CFRP
composites using a time–frequency approach", J. Intelligent Material Systems and
Structures, 23(4), pp.397-413. DOI: https://doi.org/10.1177/1045389X11434171.

[7] Bu, J.Q., Law, S.S. and Zhu, X.Q., 2006, "Innovative bridge condition assessment from
dynamic response of a passing vehicle", J. Engineering Mechanics – ASCE 132(12),
pp.1372-1379. https://doi.org/10.1061/(ASCE)0733-9399(2006)132:12(1372).

[8] McGetrick, P. J., and Kim, C. W., 2012, "Wavelet Based Damage Detection Approach
for Bridge Structures Utilising Vehicle Vibration", In Proceedings of 9th German Japanese
Bridge. Symposium, GJBS09.

[9] González, A., Obrien, E.J. and McGetrick, P.J., 2012, "Identification of damping in a
bridge using a moving instrumented vehicle", J. Sound and Vibration, 331(18), pp.4115-
4131. https://doi.org/10.1016/j.jsv.2012.04.019.

[10] Mallikarjuna Reddya, D. and Swarnamani, S., 2012, "Damage Detection and
Identification in Structures by Spatial Wavelet Based Approach", International Journal of
Applied Science and Engineering, 10(1), pp.69-87.

38
[11] Wang, HF., Noori, M., Zhao, Y., 2014, "A wavelet-based damage identification for
large crane structures", Sixth World Conference on Structural Control and Monitoring,
pp.15-17. DOI: 10.13140/2.1.2810.7207

[12] Hera, A. and Hou, Z., 2004, "Application of wavelet approach for ASCE structural
health monitoring benchmark studies", J. Engineering Mechanics, 130(1), pp.96-104.
https://doi.org/10.1061/(ASCE)0733-9399(2004)130:1(96).

[13] Gökdağ, H., Kopmaz, O., 2009, "A new damage detection approach for beam-type
structures based on the combination of continuous and discrete Wavelet Transforms", J.
Sound and Vibration, 324(3), pp.1158-1180. https://doi.org/10.1016/j.jsv.2009.02.030.

[14] Hester, D. and González, A., 2011, "A wavelet-based damage detection algorithm
based on bridge acceleration response to a vehicle", J. Mechanical Systems and Signal
Processing, 6(7), pp.145-166, https://doi.org/10.1016/j.ymssp.2011.06.007
[15] Zhao, Y., Noori, M., Altabey, W. A., Beheshti-Aval, S. B., 2017, "Mode Shape Based
Damage Identification for a Reinforced Concrete Beam Using Wavelet Coefficient
Differences and Multi-resolution Analysis", J. Structural Control & Health Monitoring,
25(1), pp. 1-41, https://doi.org/10.1002/stc.2041

[16] Zhao, Y., Noori, M., Altabey, W. A., 2017, "Damage Detection for a Beam under
Transient Excitation via Three Different Algorithms", J. Structural Engineering and
Mechanics, 64 (6), https://doi.org/10.12989/sem.2017.64.6.803

[17] Zhao, Y., Noori, M., Altabey, W. A., 2018, "A Comparison of Three Different
Methods for the Identification of Hysterically Degrading Structures Using BWBN Model",
submitted to J. Structural Control & Health Monitoring,
https://onlinelibrary.wiley.com/journal/15452263

[18] Reda Taha, M.M, Noureldin, A., Lucero, J.L. and Baca, T.J., 2006, "Wavelet
transform for structural health monitoring: a compendium of uses and features", J.
Structural Health Monitoring 5(3), pp.267–295. DOI:
https://doi.org/10.1177/1475921706067741.

[19] Nguyen, K.V. and Tran, H.T., 2010, "Multi-cracks detection of a beam-like structure
based on the on-vehicle vibration signal and wavelet analysis", J. Sound and Vibration,
329(21), pp.4455–4465. https://doi.org/10.1016/j.jsv.2010.05.005.

[20] Lee, S.G., Yun, G.J. and Shang, S., 2014, "Reference-free damage detection for truss
bridge structures by continuous relative wavelet entropy method", J. Struct. Health
Monit, 145 (2), pp.28-45. DOI: https://doi.org/10.1177/1475921714522845.

39
[21] Li, S. and Wu, Z., 2005, "Characterization of long-gauge fiber optic sensors for
structural identification, Smart Structures and Materials", International Society for
Optics and Photonics, SPIE 5765, pp.564-575. doi:10.1117/12.606367

[22] Horiguchi, T., Kurashima, T., Tateda, M., 1989, "Tensile strain dependence of
Brillouin frequency shift in silica optical fibers", Photonics Technology Letters, IEEE, 1(5),
pp.107-108. DOI: 10.1109/68.34756.

[23] Suzhen, L., and Zhishen, W., 2007, "Structural health monitoring strategy based on
distributed fiber optic sensing", J. Struct. Health Monit, 6 (2), pp.133-143. DOI:
https://doi.org/10.1177/1475921706072078.

[24] Dakin, JP., 1993, "Distributed optical fiber sensor, International Society for Optics
and Photonics", SPIE 1797, pp.76-108.

[25] Lau, KT., Yuan, L., Zhou, LM., 2001, "Strain monitoring in FRP laminates and
concrete beams using FBG sensors", J. Composite Structures, 51(1), pp.9-20.
https://doi.org/10.1016/S0263-8223(00)00094-5.

[26] Schulz, WL., Conte, JP. and Udd, E., 2002, "Real-time damage assessment of civil
structures using fiber grating sensors and modal analysis", 9th Annual International
Symposium on Smart Structures and Materials. International Society for Optics and
Photonics, SPIE 4696, pp.228-237.

[27] Wei-Xin, R., and Sun, Z., 2008, "Structural damage identification by using wavelet
entropy", J. Engineering Structures, 30(10), pp.2840-2849.
https://doi.org/10.1016/j.engstruct.2008.03.013.

[28] Ovanesova, A. V., and Suarez, L. E., 2004, "Applications of wavelet transforms to
damage detection in frame structures", J. Engineering structures, 26(1), pp.39-49.
https://doi.org/10.1016/j.engstruct.2003.08.009.

[29] Yongchao, Y., and Nagarajaiah, S., 2012, "Blind identification of damage in time-
varying system using independent component analysis with wavelet transform", J.
Mechanical Systems and Signal Processing, 27, pp.3–20.
https://doi.org/10.1016/j.ymssp.2012.08.029.

[30] Jian-Gang, H., Ren, W., and Sun, Z., 2005, "Wavelet packet based damage
identification of beam structures", International Journal of Solids and Structures 42(26),
pp.6610-6627. https://doi.org/10.1016/j.ijsolstr.2005.04.031.

[31] Yen, GG. and Lin, KC., 2000, "Wavelet packet feature extraction for vibration
monitoring", J. Industrial Electronics, IEEE Transactions on, 47(3), pp.650-667. DOI:
10.1109/41.847906.

40
[32] Sun, Z., Chang, CC., 2002, "Structural damage assessment based on Wavelet Packet
Transform", J. Structural Engineering, 128(10), pp.1354-1361.
https://doi.org/10.1061/(ASCE)0733-9445(2002)128:10(1354).

[33] Han, JG., Ren, WX., and Sun, ZS., 2005, "Wavelet packet based damage
identification of beam structures", Int. J. Solids and Structures 42(26), pp.6610-6627.
https://doi.org/10.1016/j.ijsolstr.2005.04.031.

[34] Maosen, C. and Pizhong, Q., 2008, "Integrated wavelet transform and its application
to vibration mode shapes for the damage detection of beam-type structures", J. Smart
Materials and Structures, IOP. 17(5), pp.222-232. https://doi.org/10.1088/0964-
1726/17/5/055014.

Figure Captions List

41
Fig. 1 Three level Wavelet Transform and packet transform.

Fig. 2 Flow chart of damage identification based on strain time history data.

Fig. 3 Sections of undamaged and damaged column and beam.

Fig. 4 The Numerical Models Detail.

Fig. 5 Comparison of WPER, EASC and MWPER (Normalized).

Fig. 6 Strain time history data of hammer excitation.

Fig. 7 Instantaneous strain at time 0.585s of hammer excitation.

Fig. 8 Time history data of running car excitation.

Fig. 9 Instantaneous strain at time 0.585s of running car excitation.

Fig. 10 Comparison between hammer excitation and running car excitation.

Fig. 11 The Normalized MWPER of beams.

Fig. 12 Normalized MWPER with different Contaminated Data.

Fig. 13 MWPERs of different stiffness loss levels.

Fig. 14 MWPERs of different damage locations.

Fig. 15 MPWER of different damage scenarios on the fixed ends beam.

Fig. 16 Damage index map for left front column.

Fig. 17 Section information of the tested beam.

Fig. 18 Design of the experiment.

42
 Fig. 19 Segments of all the tested beam.

Fig. 20 Hammer force time history (Test Number 5).

Fig. 21 Strain time history of Test Number 5, intact.

Fig. 22 Instantaneous strain at time=10s (Test Number 5).

Fig. 23 Damage index map for the tested beam.

Fig. 24 Damage identification result of all 12 tests.

Fig. 25 MWPER function at position 1.5m.

Table Caption List

Table 1 Section information of all the sections used in this work.

Table 2 Hammer force records.

Figures:

43
 Signal Decomposition
level

A1 D1 j=1

AA2 DA2 AD2 DD2 j=2

AAA3 DAA3 ADA3 DDA3 AAD3 DAD3 ADD3 DDD3 j=3

Discrete Wavelet transform Wavelet packet transform

Fig. 1 Three level Wavelet Transform and packet transform.

44
 Establish
Original Data
Numerical Model

Denoising
Simulate
Different Damage
Scenarios
Calculate
MWPER

Establish a
Damage Location Damage Index
Map

Damage
Damage Level
Identification

Fig. 2 Flow chart of damage identification based on strain time history data.

45
 Tt
H

D
Tb

Tl Tr
W

(a) The section used for the calculation of bending rigidity.

4
4

62
62
70

D
4

4 42 4 4 42 4
50 50

(b) Sections of undamaged and damaged beam.

46
 4
4

42
42
50

D
4

4 42 4 4 42 4
50 50

(c) Sections of undamaged and damaged column.

Fig. 3 Sections of undamaged and damaged column and beam.

47
 DB
LB=1000

(a) Simply Supported beam Model.

DB
LB=1000

(b) Fixed Ends Beam Model.

DB
LCL=LCR=1000

DCR
DCL

LB=2000

(c) Single Layer Frame Model.

Fig. 4 The Numerical Models Detail.

48
 1.0
WPER
0.9 EASC
0.8
MWPER

0.7

0.6
WPER

0.5

0.4

0.3

0.2

0.1

0.0
0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0
Location (m)

Fig. 5 Comparison of WPER, EASC and MWPER (Normalized).

49
 2.0E-3

1.5E-3

1.0E-3

5.0E-4
Strain

0.0

-5.0E-4

-1.0E-3

-1.5E-3

-2.0E-3
0.0 0.5 1.0 1.5 2.0 2.5 3.0 3.5 4.0 4.5
Time (s)

Fig. 6 Strain time history data of hammer excitation.

50
 1.5E-4

1.0E-4

5.0E-5
Strain

0.0

-5.0E-5

0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0
Location (m)

Fig. 7 instantaneous strain at time 0.585s of hammer excitation.

51
 1.0E-6

5.0E-7

0.0

-5.0E-7

-1.0E-6
Strain

-1.5E-6

-2.0E-6

-2.5E-6

-3.0E-6

-3.5E-6
0.0 0.5 1.0 1.5 2.0 2.5 3.0 3.5 4.0 4.5
Time (s)

Fig. 8 Time history data of running car excitation.

52
 2.0E-6

1.0E-6
Strain

0.0

-1.0E-6
0.0 0.5 1.0
Location (m)

Fig. 9 Instantaneous strain at time 0.585s of running car excitation.

53
 1.1
1.0
Hammer
Running Car
0.9
0.8
0.7
0.6
MWPER

0.5
0.4
0.3
0.2
0.1
0.0
-0.1
0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0
Location (m)

(a)

1.1
1.0
12.5mm
50mm
0.9
200mm
0.8
0.7
0.6
MWPER

0.5
0.4
0.3
0.2
0.1
0.0
-0.1
0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0
Location (m)

(b)
Fig. 10 Comparison between hammer excitation and running car excitation.

54
 1.0

DB=0m
0.8 DB=0.025

0.6
MWPER

0.4

0.2

0.0

0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0

Location (m)

(a) Damage at DB=0m and DB=0.025m

0.0006

0.0004
MWPER

0.0002

0.0000

0.0 0.2 0.4 0.6 0.8 1.0 1.2 1.4 1.6 1.8 2.0
Location (m)

(b) The Main Beam with Two Damages

55
 0.0002

MWPER

0.0001

0.0000

0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0
Location (m)

(c) Left Front Column

Fig. 11 The Normalized MWPER of beams.

56
 1.0
0.9
Contaminated
0.8 Uncontaminated
0.7
0.6
0.5
MWPER

0.4
0.3
0.2
0.1
0.0
-0.1
-0.2
0.0 0.5 1.0

Location (m)

(a) Contaminated data (5%).

1.0
0.9
Contaminated
0.8 Uncontaminated
0.7
0.6
0.5
MWPER

0.4
0.3
0.2
0.1
0.0
-0.1
-0.2
0.0 0.5 1.0

Location (m)

(b) Contaminated data (10%).

57
 1.0
0.9
Contaminated
0.8 Uncontaminated
0.7
0.6
0.5
MWPER

0.4
0.3
0.2
0.1
0.0
-0.1
-0.2
0.0 0.5 1.0

Location (m)

(c) Contaminated data (30%).

1.0
0.9
Contaminated
0.8 Uncontaminated
0.7
0.6
0.5
MWPER

0.4
0.3
0.2
0.1
0.0
-0.1
-0.2
0.0 0.5 1.0

Location (m)

(d) Contaminated data (50%).

Fig. 12 Normalized MWPER with different Contaminated Data.

58
 0.00030

0.00025

0.00020
MWPER

0.00015

0.00010

0.00005

0.00000
0 5 10 15 20 25 30

Stiffness loss (%)

Fig. 13 MWPERs of different stiffness loss levels.

59
 4E-5

3E-5
MWPER

2E-5

1E-5

0
0.0 0.2 0.4 0.6 0.8 1.0
Location (m)

(a) Damage Locations Spanning Across DB=0 to DB=1.

(b) The simply supported beam.

60
 (c) The fixed ends beam.

Fig. 14 MPWER of different damage scenarios

61
 3.0x10-5

2.8x10-5

2.6x10-5
MWPER

2.4x10-5

2.2x10-5

2.0x10-5

1.8x10-5
0.7 0.8 0.9 1.0 1.1 1.2 1.3 1.4 1.5 1.6
Stiffness Change

Fig. 15 MPWER of different damage scenarios on the fixed ends beam.

62
Fig. 16 Damage index map for left front column.

63
8

160
7
8

87

Fig. 17 Section information of the tested beam.

64
(a) The tested beam with sensors on both the top surface and bottom surface.

(b) Damage introduced on top surface.

Fig. 18 Design of the experiment.

65
H1 H2 H3 H4 H5 H6 H7 H8 H9 H10 H11 H12 H13

E1 E2 E3 E4 E5 E6 E7 E8 E9 E10 E11 E12

Fig. 19 Segments of all the tested beam.

66
 2500

2000

1500
Strain

1000

500

0 5 10 15 20 25 30 35 40 45 50
TIme (s)

Fig. 20 Hammer force time history (Test Number 5).

67
 0.00002

0.00001
Strain

0.00000

-0.00001

-0.00002

0 5 10 15 20 25 30 35 40 45 50
Time (s)
(a) Element 3

0.00002

0.00001
Strain

0.00000

-0.00001

-0.00002

0 5 10 15 20 25 30 35 40 45 50
Time (s)
(b) Element 5

Fig. 21 Strain time history of Test Number 5, intact.

68
 0.0 Intact beam
Damaged beam

-2.0E-6

-4.0E-6
Strain

-6.0E-6

-8.0E-6

-1.0E-5
0 5
Location (m)

Fig. 22 Instantaneous strain at time=10s (Test Number 5).

69
Fig. 23 Damage index map for the tested beam.

70
Fig. 24 Damage identification result of all 12 tests.

71
 MWPER Function
Target MWPER
0.12
MWPER

0.08

0.04

0.00
0 5 10 15 20 25 30
Stiffness loss (%)

Fig. 25 MWPER function at position 1.5m.

Tables:

Table 1 section information of all the sections used in this work.

Section Type Beam Section

Section ID 1 2 3 4 5 6 7 8

Area(mm2) 896 875 867 840 816 794 774 755

72
 Damage 0.697 1.325 1.896 2.420 2.903 3.352
0 0.5
Depth(mm) 2 2 4 2 7 5

IY(10E6mm4) 5.95 5.74 5.65 5.36 5.06 4.76 4.46 4.17

Stiffness Loss (%) 0 3.52 5 10 15 20 25 30

Table 2 Hammer force records.

Without
Test Excited With Damage
Damage
Number Point
Hammer Force(N)

1 3622.82 2184.24
H3
2 2615.97 2212.82

73
3 1907.02 2135.17

4 2480.76 1537.82

5 H5 2425.00 2093.90

6 2318.37 1883.47

7 2402.75 1883.47

8 H8 2056.68 2277.56

9 1768.33 1781.16

10 1506.53 1871.52

11 H9 1671.86 1677.18

12 1961.62 1744.82

74