You are on page 1of 20

Journal of Sound and Vibration 359 (2015) 195–214

Contents lists available at ScienceDirect

Journal of Sound and Vibration


journal homepage: www.elsevier.com/locate/jsvi

Study on an auto-correlation-function-based damage index:


Sensitivity analysis and structural damage detection
Muyu Zhang, Rüdiger Schmidt
Institute of General Mechanics, RWTH Aachen University, Templergraben 64, 52062 Aachen, Germany

a r t i c l e i n f o abstract

Article history: The damage index based on the auto correlation function to detect the damage of the
Received 12 December 2014 structure under white noise excitation is studied in detail in this paper. The maximum
Received in revised form values of the auto correlation function of the vibration response signals (displacement,
28 August 2015
velocity and acceleration) from different measurement points of the structure are col-
Accepted 4 September 2015
lected and formulated as a vector called Auto Correlation Function at Maximum Point
Handling Editor: K. Shin
Available online 26 September 2015 Value Vector (AMV), which is expressed as a weighted combination of the Hadamard
product of two mode shapes. AMV is normalized by its root mean square value so that the
influence of the excitation can be eliminated. Sensitivity analysis for the different parts of
the normalized AMV shows that the sensitivity of the normalized AMV to the local
stiffness is dependent most on the sensitivity of the Hadamard product of the two lower
order mode shapes to the local stiffness, which has a sudden change of the value around
the local stiffness change position. The sensitivity of the normalized AMV has the similar
shape and same trend that shows it is a very good damage indicator even for the very
small damage. The relative change of the normalized AMV before and after damage occurs
in the structure is adopted as the damage index to show the damage location. Several
examples of the stiffness reduction detection of a 12-story shear frame structure are
utilized to validate the results in sensitivity analysis, illustrate the effectiveness and anti-
noise ability of the AMV-based damage detection method and compare the effect of the
response type on the detectability of the normalized AMV.
& 2015 Elsevier Ltd. All rights reserved.

1. Introduction

Vibration-based structural damage detection and health monitoring is one of the most important issues related to the
safety, reliability and durability of the engineering structures, especially in mechanical, civil, aerospace and infrastructures
areas [1–4]. Effective methods and tools developed for structural health monitoring and damage detection can help not only
to prevent hazardous events, economic and human life loss, or catastrophic failures but also to prolong the service life of
these structures.
The basic idea of the vibration-based structural damage detection methods is that the structural dynamic properties (e.g.
natural frequencies, mode shapes and damping ratios) will change due to the damage-induced changes in the structural
physical properties (e.g. mass, stiffness and damping). So the changes of the structural dynamic properties imply the
changes of the structural physical properties. Therefore, the damage can be detected. The natural frequency method, which
was developed by Lifshitz and Rotem [5] in 1969, is the first dynamic parameter used for damage detection [2]. Over the past
few decades, an extensive number of research works have been conducted in the area of vibration-based damage detection.
A wide range of methodologies have been proposed, a great number of techniques have been developed and significant

http://dx.doi.org/10.1016/j.jsv.2015.09.004
0022-460X/& 2015 Elsevier Ltd. All rights reserved.
196 M. Zhang, R. Schmidt / Journal of Sound and Vibration 359 (2015) 195–214

advancement has been achieved to solve different kinds of problems encountered in all sorts of structures under a diversity
of situations. A large number of published studies used modal properties like natural frequency [6–8], mode shape [9–11],
frequency response function (FRF) [12,13], flexibility [14], impedance [15], coherence function [16] and transmissibility [17]
etc. for damage detection.
Time domain vibration response based structural damage detection methods are appealing in these years. That is
because the important health information of the structure is involved in its vibration response signals (e.g. displacement,
velocity and acceleration), which can be easily obtained by the recent conventional techniques of the vibration test. In
addition, it also has several advantages such as nondestructive, no requirement of a finite element model for the structure
and simplicity in calculation that it can be conducted real-time and online. Fassois and Sakellariou [18] classified the time-
series based damage detection methods and used three case studies of aircraft panel, a scale aircraft skeleton and a simple
nonlinear simulated structure to demonstrate the practicality and effectiveness of these methods. Hou and Noori [19]
examined the characteristic of representative vibration signals under wavelet transformation and developed a wavelet
based approach for damage detection, the usage of this method for the real acceleration data of a building response during
the 1971 San Fernando earthquake showed very promising results. Rezaei and Taheri [20] used the Empirical Modal
Decomposition (EMD) [21] technique to decompose the vibration response signals of the beams obtained from piezoceramic
sensors attached on them. The energies of the decomposed responses were successfully used for damage detection. Li and
Deng [22] combined the EMD and wavelet techniques, which can identify the time and extent of the damage of a shear
building more precisely.
Among others, some researchers used the correlation function of the vibration response for damage detection. Nichols
and Seaver [23–25] proposed the correlation-function-based transfer entropy that was used for detecting the presence of
the damage-induced nonlinearities in a thick composite sandwich plate and a composite UAV wing. Overbuy and Todd [26]
modified this transfer entropy and used it for the damage identification of an experimental 8-dof oscillator. Li and Law [27]
proposed a method based on the wavelet packet energy of cross correlation function of the measured acceleration responses
to quantify the severity of the damage. Afterwards they used the matrix of the covariance of covariance, which is also
obtained by the auto/cross correlation function of acceleration response signals under white noise excitation, successfully
detected the damages of a simply supported plane truss structure [28]. Ni and Xia [29] found the auto/cross correlation
function can be divided into two parts, one is associated with the modal parameters and the other with the excitation. They
used a two-stage method updated these two parts for detecting the damage under multiple unknown excitations
numerically and experimentally. If the damage is introduced in the structure, the dependence between the two measured
signals will decrease, thus the cross correlation between these two signals will also decrease. Taking the idea of that,
Trendafilova [30–32] developed a damage indicator from the normalized cross correlation to detect the presence and locate
the different kinds of delamination in a composite laminate beam subjected to random excitation. Yang and Yu [33] pro-
posed a vector named Cross Correlation Function Amplitude Vector (CorV). It was shown that the CorV is related to the FRF
and it has a specific shape that is useful for structural damage detection. Based on the concept of CorV and Natural Excitation
Technique (NExT) [34,35], Yang and Wang [36–39] proposed a damage detection method using the Inner Product Vector
(IPV). It was proved that the IPV is a weighted summation of the mode shapes, the effectiveness of which was demonstrated
by delamination damage detection of a simulated composite laminate beam and several damage detection experiments.
Similar to the IPV, Zhang and Schmidt [40] proposed a damage index named Auto Correlation Function at Maximum Point
Value Vector (AMV) based on the auto correlation function. Stiffness reduction detection of a 12-story frame structure
showed that using the AMV one can locate the damage effectively even when noise exist and has a better detectability
compared to the other correlation-function-based damage detection method.
In this paper, the auto-correlation-function-based damage index AMV is studied in detail. Firstly the theory of auto
correlation function of vibration responses (displacement, velocity and acceleration) is studied, based on which the damage
index is formulated. Secondly, sensitivity analysis is used to show the reason to choose this damage index for damage
detection. Damage detection of a 12-story frame structure is used to verify this AMV-based method. Comparison of the
effect of response type on its detectability is later presented. At last, some conclusions are made.

2. Damage index

2.1. Auto correlation function

The standard matrix equation for the vibration of a structure is expressed by

€ þCxðtÞ
MxðtÞ _ þKxðtÞ ¼ fðtÞ (1)

_
where M is the mass matrix, C is the damping matrix, K is the stiffness matrix, f is a vector of the random force and x(t), xðtÞ

and xðtÞ is the vector of displacement, velocity and acceleration, respectively.
M. Zhang, R. Schmidt / Journal of Sound and Vibration 359 (2015) 195–214 197

Suppose proportional damping C ¼ αM þ βK and real normal modes are assumed, the displacement xi ðtÞ, velocity x_ i ðtÞ
and acceleration x€ i ðtÞ at point i due to input f p at point p can be expressed by
Xn Z t
xi ðtÞ ¼ ψ ir ψ pr U f p ðτÞg r ðt  τÞdτ (2)
r¼1 1

X
n Z t
x_ i ðtÞ ¼ ψ ir ψ pr U f p ðτÞg_ r ðt  τÞdτ (3)
r¼1 1

X
n Z t
x€ i ðtÞ ¼ ψ ir ψ pr U f p ðτÞg€ ðt  τÞdτ
r
(4)
r¼1 1

where ψ ir , ψ pr is the ith and pth component of mode shape Ψr , respectively. And the function
(
 ζ ωrn t
sin ðωrd tÞ t Z 0
r
1
r re
g r ðtÞ ¼ m ωd (5)
0 t o0
 
is the rth modal mass, ωrn is the rth modal frequency, ζ ¼ 12 ωαr þ βωrn is the rth modal damping ratio,
r
where m qr ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi n
ωrd ¼ ωrn ½1  ðζr Þ2  is the damped modal frequency, and g_ r ðtÞ and g€ r ðtÞ is the first and second order derivative of gr ðtÞ, respectively
(  
 ζ ωrn t
 ζ ωrn sin ðωrd tÞ þ ωrd cos ðωrd tÞ t Z 0
r r
1
r re
g_ r ðtÞ ¼ m ωd (6)
0 t o0
( h i
 ζ ωrn t
ðζ ωrn Þ2 sin ðωrd tÞ  2ζ ωrn ωrd cos ðωrd tÞ  ðωrd Þ2 cos ðωrd tÞ
r r r
r
1
mr ωrd e t Z0
g€ ðtÞ ¼ (7)
0 t o0
Deduced from the definition in [41], the auto correlation function of the vibration response when the time lag T ¼0 can
be expressed as [34]
Xn X n Z 1
Ri ð0Þ ¼ αp ψ ir ψ pr ψ is ψ ps U Gr ðλÞGs ðλÞdλ (8)
r ¼1s¼1 0

where r and s are the modal order, GðtÞ is chosen according to the response type. If Ri is the displacement-response-based
auto correlation function, then GðtÞ ¼ gðtÞ as in Eq. (5). If Ri is the velocity-response-based auto correlation function, then
_
GðtÞ ¼ gðtÞ €
as in Eq. (6). And if Ri is the acceleration-response-based auto correlation function, then GðtÞ ¼ gðtÞ as in Eq. (7).
Substituting the expression of the function GðtÞ into Eq. (8) results in
" #
X
n X
n
Ri ð0Þ ¼ ψ ir ψ is μrs (9)
r¼1 s¼1

where μrs for the different response type can be expressed as follows:
Z 1
μdis
rs ¼ αp ψ pr ψ ps U g r ðλÞg s ðλÞdλ
Z0 1
1
e  ðζ ωn þ ζ ωn Þλ sin ðωrd λÞ sin ðωsd λÞdλ
r r s s
¼ αp ψ pr ψ ps U r ωr ms ωs
0 m d d
¼ I rs ϑrs (10)
Z 1
μvel
rs ¼ αp ψ pr ψ ps U g_ r ðλÞg_ s ðλÞdλ
Z0 1
1  
e  ðζ ωn þ ζ ωn Þλ
r r s s
¼ αp ψ pr ψ ps U U  ζ ωrn sin ðωrd λÞ þ ωrd cos ðωrd λÞ
r

0 mr ωrd ms ωsd
U ½  ζ ωsn sin ðωsd λÞ þ ωsd cos ðωsd λÞdλ
s
n o
¼ ½ðωrd Þ2  ðζ ωrn Þ2 I rs 2ωrd ζ ωrn J rs ϑrs
r r
(11)

Z 1
μacc
rs ¼ αp ψ pr ψ ps g€ ðλÞg€ ðλÞdλ
r s

Z 0
1
1
e  ðζ ωn þ ζ ωn Þλ
r r s s
¼ αp ψ pr ψ ps
0 mr ωrd ms ωsd
U ½ðζ ωrn Þ2 sin ðωrd λÞ 2ζ ωrn ωrd cos ðωrd λÞ  ðωrd Þ2 cos ðωrd λÞ
r r

U ½ðζ ω sin ðω λÞ  2ζ ωsn ωsd cos ðωsd λÞ ðωsd Þ2 cos ðωsd λÞdλ
s s 2 s s
nÞ d
n o
¼ 2ωrd ωrn ωsn ζ ωsn ½ðωsn Þ2 þ 4ðζ ωrn Þ2  þ ζ ωrn ½ðωrn Þ2 þ 4ðζ ωsn Þ2  ϑrs
r s s r
(12)
198 M. Zhang, R. Schmidt / Journal of Sound and Vibration 359 (2015) 195–214

where the superscripts dis, vel and acc in μrs means the auto correlation function is calculated by displacement, velocity and
acceleration response, respectively, and
αp ψ pr ψ ps
ϑrs ¼ (13)
mr ωrd ms ðI 2rs þ J 2rs Þ

I rs ¼ 2ωrd ðζ ωrn þ ζ ωsn Þ


r s
(14)

J rs ¼ ððωsd Þ2 ðωrd Þ2 Þ þ ðζ ωrn þ ζ ωsn Þ2 :


r s
(15)

2.2. Damage index

As the auto correlation function value at the time lag T ¼0 is also its maximum value, a vector named Auto Correlation
Function at Maximum Point Value Vector (AMV) [40] from n different measurement points of the structure is defined as
R AMV ¼ ½R1 ð0Þ; R2 ð0Þ; …; Rn ð0ÞT (16)
Insert Eq. (9) into Eq. (16)
2 3
n X
X n
6 μrs U ðψ 1r ψ 1s Þ 7
6 r ¼ 1s ¼ 1 7
6 7
6X 7
6 n X n
7 Xn X
6 μ U ðψ 2r ψ 2s Þ 7 n
R AMV ¼6
6 r ¼ 1s ¼ 1
rs 7¼
7 μrs UðΨr 1Ψs Þ (17)
6 7 r ¼1s¼1
6 ⋮ 7
6 n n 7
6 XX 7
4 μ U ðψ ψ Þ 5
rs nr ns
r ¼ 1s ¼ 1

where ° is the Hadamard product [42] symbol. So the vector RAMV is expressed as a weighted combination of the Hadamard
product of two mode shape vectors.
Note from Eq. (9), there is a constant αk related to the excitation in the definition of the auto correlation function.
Therefore, the vector RAMV is normalized by its root mean square value to eliminate its influence, expressed as follows:
RAMV
RAMV ¼ ¼ ½R1 ; R2 ; …; Rn T (18)
rmsðRAMV Þ
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
Pn
where rmsðRAMV Þ ¼ 1n R2i .
i¼1

Then the relative change of ith element in the vector RAMV is defined as
d u
Ri  Ri
DAMV;i ¼ u (19)
Ri

where the superscript d denotes the damaged structural state and u denotes the undamaged structural state. Thus, the
damage index of the AMV-based damage detection method is formulated as DAMV ¼ ½DAMV;1 ; DAMV ;2 ; …DAMV;n T .

3. Sensitivity analysis of the normalized AMV with respect to the local stiffness

Sensitivity analysis is widely used in the engineering fields [43] to evaluate the effect of the change of one variable on
another. One definition of the sensitivity is the limitation of the relative change of a variable x to the relative change of
another variable y when the change of y is close to zero [44,45], expressed as
Δx=x y ∂x
ηðx=yÞ ¼ lim ¼ (20)
Δy-0Δy=y x ∂y
The sensitivity of the normalized AMV in Eq. (18) with respect to the local stiffness kj is defined as
ηðRAMV =kj Þ ¼ ½ηðR1 =kj Þ; ηðR2 =kj Þ; …; ηðRn =kj ÞT (21)

where the ith element is expressed as


kj ∂Ri
ηðRi =kj Þ ¼ (22)
Ri ∂kj
using the definition of the sensitivity in Eq. (20), where Ri is the ith element in the vector RAMV in Eq. (18).
M. Zhang, R. Schmidt / Journal of Sound and Vibration 359 (2015) 195–214 199

Substitute Eq. (22) into the Hadamard product of ηðRAMV =kj Þ and R AMV results in
2 3
2 32 3 kj ∂R1
ηðR1 =kj Þ R1 6 ∂kj 7
6 76 7 6 7
6 ηðR2 =kj Þ 7 6 R2 7 6 kj ∂R2 7
ηðR AMV =kj Þ1RAMV ¼ 6 716 7 ¼ 6 ∂kj 7 ¼ kj ∂RAMV (23)
6 ⋮ 76 ⋮ 7 6 7 ∂kj
4 54 5 6 ⋮ 7
4 5
ηðRn =kj Þ Rn ∂R
kj ∂knj

Using the definition of RAMV in Eq. (18) and note


 
∂½rmsðRAMV Þ 1 1 ∂RAMV
¼ RTAMV U (24)
∂kj n rmsðRAMV Þ ∂kj

Eq. (23) can be expressed as


1 T ∂RAMV
ηðR AMV =kj Þ1RAMV ¼ ½1nn  R AMV RAMV kj (25)
n ∂kj

Define the sensitivity of μrs with respect to the local stiffness kj as ηðμrs =kj Þ
kj ∂μrs
ηðμrs =kj Þ ¼ (26)
μrs ∂kj
and the sensitivity of the Hadamard product of two mode shapes to the local stiffness kj as η½ðΨr 1Ψs Þ=kj 
   
η ðψr 1ψs Þ=kj ¼ ηðψ 1r ψ 1s =kj Þ; ηðψ 2r ψ 2s =kj Þ; …; ηðψ nr ψ ns =kj Þ T (27)

using the definition of the sensitivity in Eq. (20), the ith element of η½ðΨr 1Ψs Þ=kj 
 
k ∂ðψ ir ψ is Þ kj ∂ψ ir ∂ψ
ηðψ ir ψ is =kj Þ ¼ j ¼ ψ is þ ψ ir is (28)
ψ ir ψ is ∂kj ψ ir ψ is ∂kj kj

Using the definition of RAMV in Eq. (17), and note Eqs. (26) and (27), Eq. (25) result in

1 T
ηðRAMV =kj Þ1R AMV ¼ ½1nn  R AMV R AMV
n
Xn X n
U η½ðΨr 1Ψs Þ=kj  þ η½μrs =kj  U ½1n1 1Ψr 1Ψs U μrs (29)
r ¼1s¼1

As shown in Eq. (29), the value of ηðRAMV =kj Þ is related to the sensitivity of the Hadamard product of mode shapes to the
local stiffness kj , i.e. η½ðΨr 1Ψs Þ=kj , the sensitivity of μrs to the local stiffness kj , i.e. ηðμrs =kj Þ, the Hadamard product of mode

Fig. 1. 12-Story shear frame structure.


200 M. Zhang, R. Schmidt / Journal of Sound and Vibration 359 (2015) 195–214

shapes Ψr 1Ψs and the value of μrs . Moreover, as the coefficient of the term η½ðΨr 1Ψs Þ=kj  and ηðμrs =kj Þ are the same, their
contribution to the value of ηðR AMV =kj Þ are in the same level.

4. Sensitivity analysis results

A 12-story shear frame structure is chosen as the numerical simulation model to study the sensitivity of the normalized
AMV to the local stiffness in this paper, as shown in Fig. 1.
Assume the mass of each story is centralized on its floor with mi of each floor is 1 kg, the stiffness of each floor is supplied
by the braces between them and the stiffness in y direction is much larger than the stiffness in x direction. Therefore, only
the movement in x direction is considered and the mechanical model of this 12-story shear frame structure can be
expressed as a 12-dof discrete system. The mass matrix and the stiffness matrix of the system can be written as
M ¼ diag½m1 ; m2 ; …; m11 ; m12  (30)
2 3
k1 þ k2  k2
6 k k2 þ k3 k3 7
6 2 7
6 7
K¼6
6 ⋱ ⋱ ⋱ 7
7 (31)
6  k11 k11 þk12 k12 7
4 5
k12 k12

The stiffness coefficient ki of each floor is 20,000 N/m. Proportional damping C ¼ αM þ βK is adopted. The excitation force
is applied on the 12th floor.

4.1. Results of the sensitivity of μrs to the local stiffness

Note from the definition of μrs in Eqs. (10)–(12), μrs is a function of the frequencies and mode shape elements
μrs ¼ f ðωrn ; ωsn ; ψ pr ; ψ ps Þ (32)

So the partial derivative of μrs to the local stiffness kj is expressed by


∂μrs ∂μrs ∂ψ pr ∂μrs ∂ψ ps ∂μrs ∂ωrn ∂μrs ∂ωsn
¼ þ þ þ (33)
∂kj ∂ψ pr ∂kj ∂ψ ps ∂kj ∂ωrn ∂kj ∂ωsn ∂kj
∂ωr
where the partial derivative of the natural frequency to the local stiffness, i.e. ∂kjn and the partial derivative of the mode
Ψr are deduced from the basic vibration theories [46–48], respectively. When the propor-
shape to the local stiffness, i.e. ∂∂kj
∂ωr
tional damping is assumed, the partial derivative of the natural frequency to the local stiffness, i.e. ∂kjn can be obtained from
the real part of the following equation:
 
∂λr 1 2 ∂M ∂C ∂K
r r Ψr λr þ λr þ Ψr
T
¼ (34)
∂kj 2ðλr þ ζ ωn Þ ∂kj ∂kj ∂kj
Ψr is expressed by
where λr ¼  ζ ωrn 7 iωrd , while the partial derivative of the mode shape to the local stiffness, i.e. ∂∂k
r
j

∂Ψr Xn
¼ au Ψu (35)
∂kj u¼1

where
8  
>
<  ðλ2  λ2 Þ þ ðλ 1
ΨTu λ2r ∂M þ λr ∂k
∂C ∂K
þ ∂k Ψr ; u a r
r  λu Þ½α þ β ðωn Þ 
u 2 ∂kj j j
au ¼ r u
:
>
:  12Ψr ∂M
∂k Ψr ;
T
j
u¼r

For the 12-story frame structure used in this paper, the mass will not change when the stiffness kj changes is assumed, thus
∂M
¼0 (36)
∂kj
2 3

∂K 66 1 1 7
7
¼6 7 (37)
∂kj 4 1 1 5

where in Eq. (37) the value of 1 is on the j 1th row, j 1th column and jth row, jth column, and the value of 1 is on the
j  1th row, jth column and jth row, j 1th column. Besides as proportional damping is assumed
∂C ∂M ∂K
¼α þβ (38)
∂kj ∂kj ∂kj
M. Zhang, R. Schmidt / Journal of Sound and Vibration 359 (2015) 195–214 201

Fig. 2. Sensitivity of μdis


rs to the local stiffness kj : (a) j ¼ 2; (b) j ¼ 5; (c) j ¼ 8; and (d) j ¼ 11.

Fig. 3. Sensitivity of μvel


rs to the local stiffness kj : (a) j ¼ 2; (b) j ¼ 5; (c) j ¼ 8; and (d) j ¼ 11.
202 M. Zhang, R. Schmidt / Journal of Sound and Vibration 359 (2015) 195–214

Fig. 4. Sensitivity of μacc


rs to the local stiffness kj : (a) j ¼ 2; (b) j ¼ 5; (c) j ¼ 8; and (d) j ¼ 11.

Substituting Eqs. (36)–(38) into Eqs. (34) and (35), the partial derivative of the natural frequency and mode shape to the
local stiffness kj can be obtained, then substituting Eq. (33) into Eq. (26), the sensitivity of μrs to the local stiffness kj can be
obtained. Figs. 2–4 display the sensitivity of μdis rs (Eq. (10)), μrs (Eq. (11)) and μrs (Eq. (12)) to the local stiffness kj ,
vel acc

respectively. j is set at measurement point 2, 5, 8 and 11, respectively.


In Fig. 2, the sensitivity of μdisrs to the local stiffness kj , i.e. ηðμrs =kj Þ has a peak value when the modal order r ¼1 and s ¼1
dis

at each selected measurement point j, the other values are relatively very small. So when the local stiffness kj changes, only
the lower modal order μdis rs will change. The same phenomenon can be observed in Fig. 3 for the sensitivity of μrs to the local
vel

stiffness kj , i.e. ηðμrs =kj Þ for each selected measurement point j, while for the sensitivity of μrs to the local stiffness kj , i.e.
vel acc

ηðμacc
rs =kj Þ for each selected measurement point j shown in Fig. 4, the values follow another trend, they have relatively larger
random value when the modal order r ¼s.

4.2. Results of the sensitivity of Ψr 1Ψs to the local stiffness

Assume the damping


ratio is very small that 0 o ζ 51, then ωd  ωn .
As 0 r sin ðzÞ r 1 when z A ð  1; þ 1Þ, the expression in Eq. (10) has the range


0 r e  ðζ ωn þ ζ ωn Þλ sin ðωrd λÞ sin ðωsd λÞ r e  ðζ ωn þ ζ ωn Þλ
r r s s r r s s
(39)
R1 R1
For μdis
rs in Eq. (10), note if 0 r AðλÞ rBðλÞ, then 0 AðλÞdλ r 0 BðλÞdλ. We can obtain
Z 1
dis 1  ðζ ωrn þ ζ ωsn Þλ

sin ðωrd λÞ sin ðωsd λÞdλ
r s
μrs ¼ αp ψ pr ψ ps U se
0 mr ω
r ms
d
ω d
Z 1
1  ðζr ωrn þ ζs ωsn Þλ
r αp ψ pr ψ ps U e sin ð ω r
λÞ sin ð ωs
λÞ dλ
0 mr ω d ms ω d
r s d d

Z 1
1
e  ðζ ωn þ ζ ωn Þλ dλ
r r s s
r αp ψ pr ψ ps U r ωr ms ωs
0 m d d


ψ pr ψ ps 1
 αp r r s s r (40)
m ωn m ωn ζ ωrn þ ζ s ωsn
M. Zhang, R. Schmidt / Journal of Sound and Vibration 359 (2015) 195–214 203

Fig. 5. Value of μrs obtained from different response types when excitation force at 12th floor: (a) value of μdis
rs ; (b) value of μrs ; and (c) value of μrs .
vel acc


As 0 r cos ðzÞ r 1 when z A ð 1; þ 1Þ and 0 o ζ o o1, the expression in Eq. (11) has the range


0 r e  ðζ ωn þ ζ ωn Þλ ½  ζ ωrn sin ðωrd λÞ þ ωrd cos ðωrd λÞ U ½  ζ ωsn sin ðωsd λÞ þ ωsd cos ðωsd λÞ
r r s s r s

re  ðζ ωn þ ζ ωn Þλ ωrd ωsd
r r s s
(41)

For μvel
rs in Eq. (11), we can obtain
Z
vel
1
1
e  ðζ ωn þ ζ ωn Þλ
r r s s
μrs ¼ αp ψ pr ψ ps U
0 mr ωrd ms ωsd

U½  ζ ωrn sin ðωrd λÞ þ ωrd cos ðωrd λÞ U½  ζ ωsn sin ðωsd λÞ þ ωsd cos ðωsd λÞdλ
r s

Z 1
1  ðζr ωrn þ ζs ωsn Þλ
r αp ψ pr ψ ps U e
0 m ωd m ωd
r r s s


U½  ζ ωrn sin ðωrd λÞ þ ωrd cos ðωrd λÞ U½  ζ ωsn sin ðωsd λÞ þ ωsd cos ðωsd λÞ dλ
r s

Z 1
1
e  ðζ ωn þ ζ ωn Þλ ωrd ωsd dλ
r r s s
r αp ψ pr ψ ps U r ωr ms ωs
0 m d d


ψ pr ψ ps 1
¼ αp (42)
mr ms ζ r ωrn þ ζ s ωsn

Similar to Eq. (41), the expression in Eq. (12) has the range


0 r e  ðζ ωn þ ζ ωn Þλ U ½ðζ ωrn Þ2 sin ðωrd λÞ  2ζ ωrn ωrd cos ðωrd λÞ ðωrd Þ2 cos ðωrd λÞ
r r s s r r



U ½ðζ ωsn Þ2 sin ðωsd λÞ  2ζ ωsn ωsd cos ðωsd λÞ  ðωsd Þ2 cos ðωsd λÞ
s s

r e  ðζ ωn þ ζ ωn Þλ ðωrd Þ2 ðωsd Þ2
r r s s
(43)
204 M. Zhang, R. Schmidt / Journal of Sound and Vibration 359 (2015) 195–214

Fig. 6. Sensitivity of Ψ1 1Ψ1 , Ψ1 1Ψ2 and Ψ2 1Ψ2 to the local stiffness kj : (a) j ¼ 2; (b) j ¼ 5; (c) j ¼ 8; and (d) j ¼ 11.

For μacc
rs in Eq. (12), we can obtain
Z
acc 1
1
μ ¼ αp ψ ψ e  ðζ ωn þ ζ ωn Þλ
r r s s

rs pr ps
mr ωrd ms ωsd
0
h i
U ðζ ωrn Þ2 sin ðωrd λÞ  2ζ ωrn ωrd cos ðωrd λÞ  ðωrd Þ2 cos ðωrd λÞ
r r



U½ðζ ωsn Þ2 sin ðωsd λÞ  2ζ ωsn ωsd cos ðωsd λÞ  ðωsd Þ2 cos ðωsd λÞdλ
s s

Z 1
1  ðζr ωrn þ ζs ωsn Þλ
r αp ψ pr ψ ps e
0 m ωd m ωd
r r s s
h i
U ðζ ωrn Þ2 sin ðωrd λÞ  2ζ ωrn ωrd cos ðωrd λÞ  ðωrd Þ2 cos ðωrd λÞ
r r



U½ðζ ωsn Þ2 sin ðωsd λÞ  2ζ ωsn ωsd cos ðωsd λÞ  ðωsd Þ2 cos ðωsd λÞ dλ
s s

Z 1
1
e  ðζ ωn þ ζ ωn Þλ ðωrd Þ2 ðωsd Þ2 dλ
r r s s
r αp ψ pr ψ ps
0 mr ωrd ms ωsd


ψ pr ψ ps ωrn ωsn
 αp (44)
mr ms ζ r ωrn þ ζ s ωsn

i rh and/or s iincrease, the natural frequency ωn and/or ωn increase, then the value of
r s
When the hmodal order
ζ ωn þ ζ ωn ¼ 2 α þ βðωn Þ þ 2 α þ βðωn Þ will increase since
r r s s 1 r 2 1 s 2
the proportional
damping is assumed, so the value of
1
will decrease.
ζ r ωrn þ ζs ωsn For Eqs. (40) and (42), the value
of μdis vel
rs and μrs will decrease with the increase
of the modal order.
Moreover, μdis rs
has ωr ωs in the denominator, so μdis has a much larger decrease rate than μvel when the modal order
n n rs rs
M. Zhang, R. Schmidt / Journal of Sound and Vibration 359 (2015) 195–214 205

Fig. 7. Sensitivity of the normalized AMV to the local stiffness kj : (a) j ¼ 1; (b) j ¼ 2; (c) j ¼ 3; and (d) j ¼ 4. (For interpretation of the references to color in
this figure, the reader is referred to the web version of this article.)

Fig. 8. Sensitivity of the normalized AMV to the local stiffness kj : (a) j ¼ 5; (b) j ¼ 6; (c) j ¼ 7; and (d) j ¼ 8. (For interpretation of the references to color in
this figure, the reader is referred to the web version of this article.)

increases. For Eq. (44), value of jμacc


rs j has ωn ωn in the numerator that
r s
it cannot
be predicted
when the modal order increases,
so jμacc μdis and μvel are dependent on the lower order modal
rs j is related to all the modal
orders.
As a result, the value of rs rs
parameters, while the value of μacc rs
is dependent on all order modal parameters. The value of μ is plotted in Fig. 5. Fig. 5
rs
(a), (b) and (c) is the value of μrs , μvel
dis
rs and μrs calculated from Eqs. (10), (11) and (12), respectively.
acc
206 M. Zhang, R. Schmidt / Journal of Sound and Vibration 359 (2015) 195–214

Fig. 9. Sensitivity of the normalized AMV to the local stiffness kj : (a) j ¼ 9; (b) j ¼ 10; (c) j ¼ 11; and (d) j ¼ 12. (For interpretation of the references to color
in this figure, the reader is referred to the web version of this article.)

The results in Fig. 5 coincide well with the results obtained from Eqs. (39)–(41). For the value of μdis rs in Fig. 5(a), it has a
peak value when the modal order r ¼1 and s ¼ 1, and when the modal order r and/or s increase, the value of μdis rs decreases to
zero very fast. For the value of μvel
rs in Fig. 5(b), it has a similar trend. For the value of μacc
rs in Fig. 5(c), it has peak values when
the modal order r ¼s, and the peak values decrease when the modal order increases. As a result, only the lower order modal
parameters will affect the value of μrs . As the sensitivity of the normalized AMV to the local stiffness ηðRAMV =kj Þ has the
expression in Eq. (29), only the lower order modal parameters will affect the value of ηðRAMV =kj Þ. Therefore, only the
sensitivity of Ψ1 1Ψ1 , Ψ1 1Ψ2 and Ψ2 1Ψ2 to the local stiffness kj is considered here. The results are shown in Fig. 6, j is also
set at measurement point 2, 5, 8 and 11, respectively. The title of ‘Ψ1 1Ψ1 ’, ‘Ψ1 1Ψ2 ’ and ‘Ψ2 1Ψ2 ’ in Fig. 6(a)–(d) means that
figure is the result of the sensitivity of Ψ1 1Ψ1 , Ψ1 1Ψ2 and Ψ2 1Ψ2 to the local stiffness kj , respectively.
For each case in Fig. 6, η½ðΨr 1Ψs Þ=kj  has a sudden change of the value around the local stiffness change location, i.e. the
measurement point j, which will largely affect the value of ηðR AMV =kj Þ that discussed in later section.

4.3. Results of the sensitivity of the normalized AMV to the local stiffness

Using Eq. (29), the sensitivity of the normalized AMV to different local stiffness kj , j ¼ 1; 2; …; 12 for different response
types (displacement, velocity and acceleration) can be obtained, as shown in Figs. 7–9. For each figure, the red solid line is
the value calculated from displacement-response-based auto correlation function, the green dotted line is the value cal-
culated from velocity-response-based auto correlation function and the blue dash dotted line is the value calculated from
acceleration-response-based auto correlation function.
From Figs. 7–9, we can observe one common phenomenon of each case for the sensitivity of normalized AMV to the local
stiffness kj , i.e. ηðR AMV =kj Þ when the AMV is obtained by displacement response and velocity response that the value of
ηðRAMV =kj Þ has a sharp change around the measurement point j. Furthermore, before point j the value is positive while after
point j the value is negative, which means when there is local stiffness kj deduction in the jth floor, the value of the
normalized auto correlation function at the time lag T ¼ 0, i.e. Ri will decrease before point j, and it will increase after point j.
That is, before and after kj decreases, the relative change of Ri will change sharply around the local stiffness change location,
i.e. the measurement point j. So the local stiffness change location can be observed using the value of Ri from different
measurement points, and the damage index obtained from Eq. (19) can be used for damage localization when the response
type is chosen as displacement or velocity. For the case of the value of ηðRAMV =kj Þ when the AMV is obtained by acceleration
response in Figs. 7–9, the trend is different. It has a sharp change around the local stiffness change location, and for the other
points, it is normally the same. As a result, the relative change of Ri before and after kj decreases is normally the same before
and after the stiffness change location, i.e. measurement point j, and it has a highest relative change rate at measurement
point j. So using the value of Ri from different measurement points, local stiffness change location can also be observed, the
M. Zhang, R. Schmidt / Journal of Sound and Vibration 359 (2015) 195–214 207

Fig. 10. Value of μrs obtained from different response types when excitation force at 1st floor: (a) value of μdis
rs ; (b) value of μrs ; and (c) value of μrs .
vel acc

damage index calculated from Eq. (19) can also be used for damage localization when the response type is chosen as
acceleration.
The shape and trend of the sensitivity analysis results for each case in Figs. 7–9 can be explained using Eq. (29). The value
of ηðR AMV =kj Þ is dependent on the value of μrs, Ψr 1Ψs , ηðμrs =kj Þ and η½ðΨr 1Ψs Þ=kj . Compare the values of ηðμrs =kj Þ in Figs. 2–4
and the values of η½ðΨr 1Ψs Þ=kj  in Fig. 6, ηðμrs =kj Þ is far smaller than η½ðΨr 1Ψs Þ=kj  for each case. As from the expression in Eq.
(29) the contribution of ηðμrs =kj Þ and η½ðΨr 1Ψs Þ=kj  to the value of ηðRAMV =kj Þ are in the same level, so the value of
η½ðΨr 1Ψs Þ=kj  contribute far more than the value of ηðμrs =kj Þ to the value of ηðRAMV =kj Þ. Besides, as different elements in
Ψr 1Ψs are in the same level, the value of ηðRAMV =kj Þ depends most on the value of η½ðΨr 1Ψs Þ=kj  and the value of μrs .
Furthermore, as the value of μrs drops to close to zero sharply when the modal order increases, as shown in Fig. 5, the value
of ηðR AMV =kj Þ is only related to the lower modal order value of η½ðΨr 1Ψs Þ=kj . As a result, as there is a sharp change around
the local stiffness change location for the lower modal order value of η½ðΨr 1Ψs Þ=kj , as shown in Fig. 6, the value of
ηðRAMV =kj Þ also have a sharp change around the local stiffness change location that are shown in Figs. 7–9.
As to the difference between the displacement- and velocity-response-based ηðRAMV =kj Þ and acceleration-response-
based ηðRAMV =kj Þ, this is because of the value of μrs . From Section 4.2, the values of μdis rs and μrs have the same trend, they
vel

both have peak values when the modal order r and s are relatively very small compare to the value of μacc rs . So the dis-
placement- and velocity-response-based ηðRAMV =kj Þ dependent mostly on the relatively smaller modal order value of
η½ðΨr 1Ψs Þ=kj , that makes them similar to the trend of η½ðΨ1 1Ψ1 Þ=kj , while the acceleration-response-based ηðR AMV =kj Þ has
the trend of the combination of the several lower modal order values of η½ðΨr 1Ψs Þ=kj .

4.4. Effect of the excitation position on the sensitivity analysis results

From Sections 4.1 to 4.3 we can see that the value of μrs takes a very important part to determine the shape and trend of
the value of ηðRAMV =kj Þ. As it can be seen from the expression of μrs in Eqs. (10)–(12), μrs is also related to the excitation point
p. In Sections 4.1–4.3 all the cases are for the excitation point at the 12th floor, which is near the tip of the structure.
If the excitation point is near the end of the structure, i.e. at the 1st floor, the value of μrs is shown in Fig. 10, in which
(a) is the value of μdis of μrs and
rs , (b) is the value
vel
(c) is the value of μrs . The value of μrs also coincide
acc
well
with the results
obtained from Eqs. (39) to (44) that μdis rs
and μvel decrease when the modal order increases and μacc is dependent on all
rs rs
the modal orders. Comparing Figs. 5(a) and 10(a) we can see that the shape and the trend of the value of μdis rs barely changes,
so the value of the displacement-response-based ηðRAMV =kj Þ will not change much when the excitation point changes. As a
result, the detectability of the displacement-response-based AMV is all the same for the different excitation points.
208 M. Zhang, R. Schmidt / Journal of Sound and Vibration 359 (2015) 195–214

But this is not the case for the value of μvel


rs and μrs . The shape and trend of μrs in Fig. 10(b) is not similar to Fig. 5(b), on
acc vel

the other hand, it is more similar to Fig. 5(c) that has peak values when the modal order r ¼ s and the peak values decrease
when the modal order increases. So the shape and trend of the velocity-response-based ηðRAMV =kj Þ when the excitation
point is located at the 1st floor is similar to the shape and trend of the acceleration-response-based ηðRAMV =kj Þ when the
excitation point is at 12th floor. As a result, the detectability of the velocity-response-based AMV will change when the
excitation point changes. However, it can still be used for damage detection. For the value of μaccrs in Fig. 10(c), it also has peak
values when the modal order r ¼ s, but the peak values increase when the modal order increases, which is opposite as in
Fig. 5(c). So the value of ηðRAMV =kj Þ will be affected more by the higher modal order values of η½ðΨr 1Ψs Þ=kj , which do not
have the sharp change around the local stiffness change location that makes the normalized AMV can’t locate the damage.
As a result, the detectability of the acceleration-response-based AMV becomes worse when the excitation point moves from
the tip to the end of the structure.

5. Damage detection using the normalized AMV

In this section, stiffness reduction detection of the 12-story shear frame structure shown in Fig. 1 is used as the simu-
lation example to verify the AMV-based damage index. The first 12 natural frequencies of the structure can be easily
obtained when the mass matrix M in Eq. (30) and stiffness matrix K in Eq. (31) are known, they are 2.83, 8.44, 13.91, 19.17,
24.12, 28.69, 32.82, 36.42, 39.45, 41.85, 43.60 and 44.66 Hz. Stiffness coefficient of the floor 2, 5, 8 and 11 is 5% reduced as the
damage, respectively. The corresponding different damage cases are listed in Table 1. When the damage appears in the
structure, the relative change of the first natural frequency compared to the undamaged structure is also shown in Table 1.
As can be seen from Table 1, the relative change of the first natural frequency is less than 0.5%. Since the simulated damage
in this paper is very small, the natural frequency is not a good indicator for the damage.
As the white noise excitation that covers all the frequencies is used in the deduction of the auto correlation function in
Section 2.1, the white noise with a frequency range of 0–50 Hz that covers all the first 12 frequencies of the 12-story frame

Table 1
Damage cases of the frame structure.

Damage case D2 D5 D8 D11


Damaged floor 2 5 8 11
Stiffness coefficient reduction/% 5
Frequency change (%)  0.40  0.30  0.15  0.03

u d
Fig. 11. Damage detection results of damage case D2: (a) before damage: normalized AMV Ri ; (b) after damage: normalized AMV Ri ; (c) damage index;
and (d) damage location index. (For interpretation of the references to color in this figure, the reader is referred to the web version of this article.)
M. Zhang, R. Schmidt / Journal of Sound and Vibration 359 (2015) 195–214 209

u d
Fig. 12. Damage detection results of damage case D5: (a) before damage: normalized AMV Ri ; (b) after damage: normalized AMV Ri ; (c) damage index;
and (d) damage location index.

u d
Fig. 13. Damage detection results of damage case D8: (a) before damage: normalized AMV Ri ; (b) after damage: normalized AMV Ri ; (c) damage index;
and (d) damage location index. (For interpretation of the references to color in this figure, the reader is referred to the web version of this article.)

structure is adopted as the excitation in this section. The excitation has a sample frequency of 1024 Hz with duration of 16 s
and magnitude of 1 N, which is applied on the 12th floor of the frame structure. Different types of responses from the
undamaged structure and different damage cases are obtained from the Wilson-θ method.
The value of the auto correlation function can be easily calculated using the inner product [49] of the responses. After the
auto correlation function of the responses before and after damage are obtained, the damage index can be calculated by Eqs.
210 M. Zhang, R. Schmidt / Journal of Sound and Vibration 359 (2015) 195–214

u d
Fig. 14. Damage detection results of damage case D11: (a) before damage: normalized AMV Ri ; (b) after damage: normalized AMV Ri ; (c) damage index;
and (d) damage location index. (For interpretation of the references to color in this figure, the reader is referred to the web version of this article.)

(18) and (19). In order to make the damage more clearly to be seen, the AMV-based damage location index is defined by
D0AMV ;i þ 0:5 ¼ DAMV;i þ 1  DAMV;i (45)

where the local maxima of D0 corresponding to the abrupt change in the damage index D. The abrupt change of the damage
index is considered as a result of the damage occurring in the structure. Then the damage is located between the mea-
surement points i and iþ1 if the local maxima of D0 appears in the measurement point iþ0.5.

5.1. Damage detection results

Figs. 11–14 show the damage detection results of the four different damage cases using different response type. For each
figure in Figs. 11–14, the upper left named (a) is the value of the normalized AMV before damage, the upper right named
(b) is the value of the normalized AMV after damage, the lower left named (c) is the damage index and the lower right
named (d) is the damage location index. Compare each figure (a) and figure (b) in Figs. 11–14, the change of the normalized
AMV before and after damage can be hardly noticed. But through the difference between them, the damage location
appears. For each figure, the red solid line is the value calculated using displacement responses, the green dotted line is the
value calculated using velocity responses and the blue dash dotted line is the value calculated using acceleration responses.
The damage indexes in Figs. 11(c), 12(c), 13(c) and 14(c) coincide well with the sensitivity analysis results in Section 4.3.
Take the red solid line in Figs. 8(d) and 13(c) for example. The red solid line in Fig. 8(d) is the sensitivity analysis result of the
normalized AMV calculated from displacement-response-based auto correlation function when the local stiffness change
location is set at the 8th floor, the red solid line in Fig. 13(c) is the value of damage index calculated by displacement
responses when the 8th floor has a 5% stiffness reduction. In Fig. 8(d), the value of ηðRi =k8 Þ is positive from measurement
point 1 to measurement point 7 and they are almost the same, but it drops sharply at measurement point 8, where it is
negative. After that, from measurement point 8 to measurement point 12, the value slowly increases to another level again.
As a result, ηðRi =k8 Þ has the minimal value at measurement point 8. That is, when the local stiffness in the 8th floor, i.e. k8
decreases, R7 will decrease while R8 will increase, and the difference between the decrease rate and increase rate is very big.
Thus, before and after the value of k8 decreases, the relative change of R7 (means the damage index at measurement point
7), i.e. DAMV ;7 has a negative value while the relative of change of R8 (means the damage index at measurement point 8), i.e.
DAMV;8 has a positive value and there is a large difference between DAMV;7 and DAMV;8 , which can be seen in Fig. 13(c).
Moreover, for the value of ηðRi =k8 Þ, from measurement point 1 to measurement point 7 it is almost the same and from
measurement point 8 to measurement point 12 it just has a small difference. So the relative change of R1 –R7 and the relative
change of R8 –R12 before and after k8 decreases each just has a very slight change. That is, the value of DAMV;1 –DAMV;7 as well
as DAMV ;8 –DAMV;12 do not change much, which can be also seen in Fig. 13(c). The other measurement points and cases have
the same phenomenon and can be explained in the similar way. As the sensitivity analysis results in Section 4.3 only uses
the modal parameters (frequency and mode shape) of the structure and the damage detection results in this section only
M. Zhang, R. Schmidt / Journal of Sound and Vibration 359 (2015) 195–214 211

Fig. 15. Displacement-response-based damage location index under different SNRs of noises: (a) damage case D2; (b) damage case D5; (c) damage case D8;
and (d) damage case D11.

uses the time domain vibration responses (displacement, velocity and acceleration) of the structure, their well agreement
shows the normalized AMV studied in this paper is a good damage indicator to detect the damage.
The damage location index of the AMV method is plotted in Figs. 11(d), 12(d), 13(d) and 14(d), in which all these four
cases show the damage location correctly. The AMV-based damage location index has a peak value at point 1.5 in Fig. 11(d),
which indicates the damage occurs between measurement point 1 and measurement point 2, that is, the damage is in the
2nd floor. Similarly, there is a peak value at point 4.5, 7.5 and 10.5 in Figs. 12(d), 13(d) and 14(d) respectively, which indicates
the damage occurs at the 5th, the 8th and the 11th floor of the structure respectively. The damage locations detected by the
AMV-based damage location index are just the four damage cases simulated in Table 1. So the AMV method is effective in
detecting the damage.
From the results in Figs. 11–14, the damage location index calculated from displacement and velocity responses are
normally the same, but the damage location index calculated from acceleration responses follow another trend. Although it
can also locate the damage, it has false positive in Figs. 12(d) and 13(d). This is because the different types of abrupt changes
of the damage index, for the displacement- and velocity-response-based damage index it is a ‘step change’ [35] while for the
acceleration-response-based damage index it is an ‘impulse change’ [35]. For the ‘step change’, the difference of the damage
index in Eq. (42), one sudden change of the damage index gives just one peak value of the damage location index that makes
the damage more clearly to be detected, but this is not the case for the ‘impulse change’ that one sudden change of the
damage index result in two peak values of the damage location index which gives false positive. As a result, it is better to use
the damage index instead of the damage location index to locate the damage when the acceleration response is used.
Besides, the acceleration-response-based AMV method also has advantage compared to the displacement- and velocity-
response-based AMV method. It has a much larger peak value around the damage location when the damage is small, as
shown in Figs. 13 and 14. So it has a better detectability of the small damage.

5.2. Damage detection using the response with measurement noise

The measurement noise will affect the response of the structure, which is very common in real application. So in this
section noise is added to the response signals to verify the anti-noise ability of the AMV-based damage detection method.
Suppose the measurement noise simulated here is Gaussian noise and the Signal to Noise Ratio (SNR) [50] is defined as

Pox
SNR ðdBÞ ¼ 10log10 (46)
PoN

where Pox is the power of the signal and PoN is the power of the noise. Noises with three different SNRs, i.e. 30 dB (noise
212 M. Zhang, R. Schmidt / Journal of Sound and Vibration 359 (2015) 195–214

Fig. 16. Velocity-response-based damage location index under different SNRs of noises: (a) damage case D2; (b) damage case D5; (c) damage case D8;
and (d) damage case D11.

Fig. 17. Acceleration-response-based damage location index under different SNRs of noises: (a) damage case D2; (b) damage case D5; (c) damage case D8;
and (d) damage case D11.
M. Zhang, R. Schmidt / Journal of Sound and Vibration 359 (2015) 195–214 213

level 3.2%), 20 dB (noise level 10.0%) and 10 dB (noise level 31.6%) are added to the different response signals, respectively.
For each SNR, response type and structural health state, 300 sets of simulations are performed.
The mean value of the AMV-based damage location index from different SNRs, response types and structural health
states calculated from each 300 sets of simulations are plotted in Figs. 15–17. Fig. 15 shows the damage detection results
using the displacement responses, Fig. 16 shows the damage detection results using the velocity responses and Fig. 17 shows
the damage detection results using the acceleration responses. For each figure (a), (b), (c) and (d), they are the results for
structural health states D2, D5, D8 and D11, respectively. For each structural health state, the damage location index cal-
culated from the vibration responses with no noise, 30 dB noise, 20 dB noise and 10 dB noise are plotted in one figure for
comparison. As shown in these figures, even when the SNR of the noise is as low as 10 dB, there is nearly no change or just
some very small fluctuations of the damage location index compared to the results when there is no noise, which the
damage can be correctly located using the way in Section 5.1. As a result, the AMV-based damage location index can locate
the damage for each structural health state, response type and SNR, which shows the anti-noise ability of the method is very
good and therefore can be used in real application.
Similar to the results in Section 5.1, the displacement- and velocity-response-based AMV have no false positive while the
acceleration-response-based AMV gives false positives in Fig. 17(b) and (c) when the measurement noise exists. On the other
hand, compared to the damage detection results in Figs. 15–17, the fluctuations of the damage location index is the smallest
for the acceleration-response-based AMV and the largest for the displacement-response-based AMV when there is noise in
the response signals. So the measurement noise will affect the detectability of the acceleration-response-based AMV least
and displacement-response-based AMV most.

6. Conclusions

In this paper, the damage index based on the auto correlation function named AMV is studied in detail. The displace-
ment-, velocity- and acceleration-response-based AMV is expressed as a weighted combination of the Hadamard product of
two mode shapes, which can be directly obtained using the structural vibration response signals. Sensitivity analysis results
show that the sensitivity of the normalized AMV to the local stiffness is dependent most on the sensitivity of the Hadamard
product of the two lower modal order mode shapes to the local stiffness, which has a sharp change around the local stiffness
change location that is suitable for damage localization. Damage detection results of a 12-story frame structure prove the
results in sensitivity analysis and show the displacement-, velocity- and acceleration-response-based AMV can all locate the
simulated damages even when the low SNR noise exists. Although the acceleration-response-based AMV gives some false
positives, it has its own advantages that it has the best damage detection results for the small damage and it has the best
anti-noise ability.
The AMV-based damage detection method uses only the vibration response signals of the structure under white noise
excitation. While it requires no finite element model or modal analysis, it can be easily conducted in real time and online,
which is suitable for the health monitoring of the structure. As displacement-, velocity- and acceleration-response-based
AMV all can locate the damage well, one can choose the most convenient way to acquire the response signal for damage
detection.

Acknowledgments

The first author would gratefully acknowledge the financial support from the China Scholarship Council with the No.
2011629074. The authors are also grateful to Dr. Zhichun Yang, Dr. Le Wang and Dr. Michael Ban for their kind help and
discussion.

References

[1] C.R. Farrar, K. Worden, An introduction to structural health monitoring, Philosophical Transactions of the Royal Society A 365 (2007) 303–305.
[2] S.W. Doebling, C.R. Farrar, M.B. Prime, D.W.Shevitz, Damage identification and health monitoring of structural and mechanical systems from changes
in their vibration characteristics: a literature review, Los Alamos National Laboratory Report LA-13070-MS, 1996.
[3] H.S. Sohn, C.R. Farrar, F.M. Hemez, D.D. Shunk, D.W. Stinemates, B.R.Nadler, A review of structural health monitoring literature: 1996–2001, Los Alamos
National Laboratory Report LA-13976-MS, 2003.
[4] D. Montalvao, N.M.M. Maia, A.M.R. Rbeiro, A review of vibration-based structural health monitoring with special emphasis on composite materials,
The Shock and Vibration Digest 38 (2006) 295–324.
[5] J.M. Lifshitz, A. Rotem, Determination of reinforcement unbonding of composites by a vibration technique, Journal of Composite Materials 3 (1969)
412–423.
[6] E. Görl, M. Link, Damage identification using changes of eigenfrequencies and mode shapes, Mechanical Systems and Signal Processing 17 (1) (2003)
103–110.
[7] M. Dilena, A. Morassi, Vibrations of steel-concrete composite beams with partially degraded connection and applications to damage detection, Journal
of Sound and Vibration 320 (2009) 101–124.
[8] J.T. Kim, Y.S. Ryu, H.M. Cho, N. Stubbs, Damage identification in beam-type structures: frequency-based method vs mode-shape-based method,
Engineering structures 25 (1) (2003) 57–67.
214 M. Zhang, R. Schmidt / Journal of Sound and Vibration 359 (2015) 195–214

[9] A. Messina, E.J. Williams, T. Contursi, Structural damage detection by a sensitivity and statistical-based method, Journal of Sound and Vibration 216
(1998) 791–808.
[10] A.K. Pandey, M. Biswas, M.M. Samman, Damage detection from changes in curvature mode shapes, Journal of Sound and Vibration 145 (2) (1991)
321–332.
[11] D.J. Ewins, Modal Testing: Theory and Practice, Wiley, New York, 1985.
[12] C. Zang, M. Imregun, Structural damage detection using artificial neural networks and measured FRF data reduced via principal component projection,
Journal of Sound and Vibration 242 (5) (2001) 813–827.
[13] N.M.M. Maia, J.M.M. Silva, E.A.M. Almas, R.P.C. Sampaio, Damage detection in structures: from mode shape to frequency response function methods,
Mechanical Systems and Signal Processing 17 (3) (2003) 489–498.
[14] A.K. Pandey, M. Biswas, Damage detection in structures using changes in flexibility, Journal of Sound and Vibration 169 (1994) 3–17.
[15] G. Park, H. Sohn, C.R. Farrar, D.J. Inman, Overview of piezoelectric impedance-based health monitoring and path-forward, The Shock and Vibration
Digest 35 (2003) 451–463.
[16] S.S. Wang, Q.W. Ren, Structure damage detection using local damage factor, Journal of Vibration and Control 12 (9) (2006) 955–973.
[17] N.M.M. Maia, R.A.B. Almeida, A.P.V. Urguerira, R.P.C. Sampaio, Damage detection and quantification using transmissibility, Mechanical Systems and
Signal Processing 25 (2011) 2475–2483.
[18] S.D. Fassois, J.S. Sakellariou, Time-series methods for fault detection and identification in vibrating structures, Philosophical Transactions of the Royal
Society A: Mathematical, Physical and Engineering Sciences 365 (1851) (2007) 411–448.
[19] Z. Hou, M. Noori, R.S. Amand, Wavelet-based approach for structural damage detection, Journal of Engineering Mechanics 126 (7) (2000) 677–683.
[20] D. Rezaei, F. Taheri, Damage identification in beams using empirical mode decomposition, Structural Health Monitoring 10 (3) (2011) 261–274.
[21] N.E. Huang, Z. Shen, S.R. Long, et al., The empirical mode decomposition and the Hilbert spectrum for nonlinear and non-stationary time series
analysis, Proceedings of the Royal Society of London. Series A: Mathematical, Physical and Engineering Sciences 454 (1971) (1998) 903–995.
[22] H.L. Li, X.Y. Deng, H.L. Dai, Structural damage detection using the combination method of EMD and wavelet analysis, Mechanical Systems and Signal
Processing 21 (2007) 298–306.
[23] J.M. Nichols, M. Seaver, S.T. Trickey, et al., Detecting nonlinearity in structural systems using the transfer entropy, Physical Review E 72 (4) (2005)
046217.
[24] J.M. Nichols, M. Seave, S.T. Trickey, A method for detecting damage-induced nonlinearities in structures using information theory, Journal of Sound and
Vibration 297 (1) (2006) 1–16.
[25] J.M. Nichols, M. Seaver, S.T. Trickey, L.W. Salvino, D.L. Pecora, Detecting impact damage in experimental composite structures: an information-
theoretic approach, Smart Materials and Structures 15 (2) (2006) 424–434.
[26] L.A. Overbey, M.D. Todd, Dynamic system change detection using a modification of the transfer entropy, Journal of Sound and Vibration 322 (1) (2009)
438–453.
[27] X.Y. Li, S.S. Law, Condition assessment of structures under ambient white noise excitation, AIAA Journal 46 (6) (2008) 1395–1404.
[28] X.Y. Li, S.S. Law, Matrix of the covariance of covariance of acceleration responses for damage detection from ambient vibration measurements,
Mechanical Systems and Signal Processing 24 (2010) 945–956.
[29] P. Ni, Y. Xia, S.S. Law, et al., Structural damage detection using auto/cross-correlation functions under multiple unknown excitations, International
Journal of Structural Stability and Dynamics 14 (5) (2014) 1440006.
[30] I. Trendafilova, A method for vibration-based structural interrogation and health monitoring based on signal cross-correlation, Journal of Physics:
Conference Series 305 (2011) 012005.
[31] I. Trendafilova I, R. Palazzetti, A. Zucchelli, Delamination assessment in structures made of composites based on general signal correlation, Interna-
tional Journal of Structural Stability and Dynamics (2014).
[32] I. Trendafilova, R. Palazzetti, A. Zucchelli, Damage assessment based on general signal correlation. Application for delamination diagnosis in composite
structures, European Journal of Mechanics-A/Solids 49 (2015) 197–204.
[33] Z.C. Yang, Z.F. Yu, H. Sun, On the cross correlation function amplitude vector and its application to structural damage detection, Mechanical Systems and
Signal Processing 21 (2007) 2918–2932.
[34] G.H. James III, T.G. Carne, J.P. Lauffer, The natural excitation technique (NExT) for modal parameter extraction from operating structures, Modal
Analysis: The International Journal of Analytical and Experimental Modal Analysis 10 (4) (1995) 260–277.
[35] G.H. James III, C.R. Farrar, System identification from ambient vibration measurements on a bridge, Journal of Sound of Vibration 205 (1) (1997) 1–18.
[36] Z.C. Yang, L. Wang, H. Wang, Y. Ding, X.J. Dang, Damage detection in composite structures using vibration response under stochastic excitation, Journal
of Sound and Vibration 325 (2009) 755–768.
[37] L. Wang, Z.C. Yang, T.P. Waters, Structural damage detection using cross correlation functions of vibration response, Journal of Sound and Vibration 329
(2010) 5070–5086.
[38] L. Wang, Z.C. Yang, T.P. Waters, M.Y. Zhang, Theory of inner product vector and its application to multi-location damage detection, Journal of Physics:
Conference Series 305 (2011) 012003.
[39] L. Wang, Z.C. Yang, Effect of response type and excitation frequency range on the structural damage detection method using correlation functions of
vibration responses, Journal of Sound and Vibration 332 (2013) 645–653.
[40] M. Zhang, R. Schmidt, Sensitivity analysis of an auto-correlation-function-based damage index and its application in structural damage detection,
Journal of Sound and Vibration 333 (26) (2014) 7352–7363.
[41] J.S. Bendat, A. Piersol, Random Data: Analysis and Measurement Procedures, fourth ed. J. Wiley.
[42] G.P.H. Styan, Hadamard products and multivariate statistical analysis, Linear Algebra and its Applications 6 (1973) 217–240.
[43] A. Saltelli, M. Ratto, T. Andres, Global Sensitivity Analysis: The Primer, John Wiley & Sons, 2008.
[44] P. Reichert, Concepts Underlying a Computer-program for the Identification and Simulation of Aquatic Systems, Schriftenreihe der EAWAG, No. 7, EAWAG,
Duebendorf, Switzerland, 1994.
[45] P.H. Brunner, H. Rechberger, Practical handbook of material flow analysis, The International Journal of Life Cycle Assessment 9 (5) (2004) 337–338.
[46] S. Adhikari, Structural Dynamic Analysis with Generalized Damping Models: Identification, Wiley-ISTE, 2013.
[47] I.W. Lee, D.O. Kim, G.H. Jung, Natural frequency and mode shape sensitivities of damped systems: part I, distinct natural frequencies, Journal of Sound
and Vibration 223 (3) (1999) 399–412.
[48] K.M. Choi, H.K. Jo, W.H. Kim, I.W. Lee, Sensitivity analysis of non-conservative eigensystems, Journal of Sound and Vibration 274 (3) (2004) 997–1011.
[49] S.J. Orfanidis, Optimum Signal Processing: An Introduction, second ed. Prentice-Hall, Englewood Cliffs, NJ, 1996.
[50] S. Elliott, Signal Processing for Active Control, Academic Press, 2000.

You might also like