Download as pdf or txt
Download as pdf or txt
You are on page 1of 25

STRUCTURAL CONTROL AND HEALTH MONITORING

Struct. Control Health Monit. (2012)


Published online in Wiley Online Library (wileyonlinelibrary.com). DOI: 10.1002/stc.1486

A new approach to system identification and damage detection of


high-rise buildings

M. Kuwabara, S. Yoshitomi and I. Takewaki*,†


Department of Architecture and Architectural Engineering, Kyoto University, Kyotodaigaku-Katsura, Nishikyo-ku, Kyoto
615-8540, Japan

SUMMARY
A new method of system identification and damage detection of high-rise buildings is proposed in which high-rise
buildings are represented by a shear-bending model. The method is devised to find the story shear and bending stiffnesses
of a specific story from the floor accelerations just above and below the specific story. The special characteristic of the
proposed method is to derive a set of closed-form expressions for the story shear and bending stiffnesses in terms of
the limited floor accelerations. It is demonstrated that the method is suitable for the damage detection because the
observation of floor accelerations at two stories is a simple task and damage detection can be performed directly from
the change of a specific function of frequency at zero frequency. It is shown that a difficulty of prediction of the specific
function in a low frequency range can be overcome by introducing an ARX (Auto-Regressive model with eXogenous
input) model with constraints on transfer functions. Copyright © 2012 John Wiley & Sons, Ltd.

Received 18 October 2011; Revised 29 December 2011; Accepted 24 January 2012

KEY WORDS: system identification, damage detection, high-rise building, shear-bending model, ARX model

1. INTRODUCTION
Much interest has been directed recently to system identification in response to the increasing need of
damage detection (or damage diagnosis) of building structures under or after earthquakes [1–23]. Such need
of damage detection results from the accelerated demand of rapid assessment for continuing use of
buildings after earthquakes. A comprehensive review of system identification and damage detection has been
made in [21]. It is also well recognized that system identification plays an important role in reducing gaps
between the constructed structural systems and their structural design models (model refinement).
Modal parameter system identification is well established, and a versatile research has been
accumulated (for example, [1,2,24]). In the modal parameter system identification, observation at
two places is necessary for natural frequency and damping ratio identification. On the other hand,
observation (or interpolation from fewer observations) at many places is usually required for modal shape
identification. This is often a cumbersome task.
In contrast to such modal parameter system identification, physical parameter system identification has
been developed for direct identification of physical parameters (stiffness and damping coefficients). For
example, Takewaki and Nakamura [25,26] introduced a method based on the work by Udwadia et al.
[27]. In that method, a shear building model is used and stiffness and damping coefficients of a specific
story are identified directly from the floor accelerations just above and below the specific story. However,
the method by Takewaki and Nakamura [25,26] has a difficulty resulting from the small signal/noise (SN)
ratio in the low frequency range and cannot be applied to high-rise buildings with large aspect ratios.

*Correspondence to: Izuru Takewaki, Department of Architecture and Architectural Engineering, Kyoto University,
Kyotodaigaku-Katsura, Nishikyo-ku, Kyoto 615-8540, Japan.

E-mail: takewaki@archi.kyoto-u.ac.jp

Copyright © 2012 John Wiley & Sons, Ltd.


M. KUWABARA, S. YOSHITOMI AND I. TAKEWAKI

Hernandez-Garcia et al. [28,29] have developed an interesting method of damage detection using a
floor-by-floor approach to enhance the efficiency and accuracy of the identification results. Furthermore,
a combined method of the modal parameter system identification and the physical parameter system
identification is also well used (for example, [9,30]). After modal parameters are identified, physical
parameters are determined by solving inverse problems.
The difficulty arising in the limit manipulation in the method by Takewaki and Nakamura [25,26] is
overcome by introducing an ARX model in this paper. The weakness of a small SN ratio in the low
frequency range in the method [25,26] is avoided by using the ARX model and introducing new
constraints on the ARX parameters. Another difficulty due to large aspect ratios is tackled by introducing
a shear-bending model. It is shown that the proposed system identification method is applicable to damage
detection problems and the proposed damage detection method includes only a simple manipulation. The
validity of the method is examined through numerical examples.

2. GOVERNING EQUATIONS
2.1. Shear-bending model and equations of motion
Consider an N-story shear-bending building model as shown in Figure 1. Every story of this shear-bending
model consists of a rotational and extensional springs. These two springs are connected in series. Let ksj and
kbj denote the stiffness of the extensional and rotational springs, respectively. Similarly, let csj and cbj
denote the damping coefficient of the extensional and rotational dashpots, respectively. These two
dashpots are also connected in series as in the springs. The floor mass and its rotational mass moment
of inertia are denoted by mj and Jj, respectively. The height of the jth story is given by Hj.
Let [M], [K], and [C] denote the mass, stiffness, and damping matrices of this shear-bending building
model, and let yj and θj denote the horizontal floor displacement and the angle of floor rotation,
respectively. The set of floor displacements and angles of rotation is denoted by {x}. The equations of
motion for this model subjected to a horizontal ground acceleration üg can be expressed as

½M f€x g þ ½C fx_ g þ ½K fxg ¼ ½M frg€u g ; (1)


ug + yN

ug + y j

u g + y2

u g + y1

ug

Figure 1. Shear-bending model.

Copyright © 2012 John Wiley & Sons, Ltd. Struct. Control Health Monit. (2012)
DOI: 10.1002/stc
SYSTEM IDENTIFICATION AND DAMAGE DETECTION OF HIGH-RISE BUILDINGS

where
½M  ¼ diagð m1 ; ⋯; mN J1 ; ⋯; JN Þ

fxg ¼ f y1 ; ⋯; yN j θ1 ; ⋯; θN gT
()
fr g ¼ f 1; ⋯; 1 j 0; ⋯; 0 gT (2a–e)

2.2. Dynamic equilibrium in frequency domain


Let vj and fj denote the shear (elongation of the extensional spring) and bending deformations (angle
of rotation of the rotational spring), respectively. Then, vj and fj can be expressed in terms of yj and θj.
vj ¼ yj  yj1  θj Hj (3a)
fj ¼ θj  θj1 (3b)
From the dynamic equilibrium of a free body shown in Figure 2, the following equations can be derived.

− mN (u g + y N )

− m j +1 (u g + y j +1 )

− m j (u g + y j )

k sj v j + csj v j

kbj j + cbj j
− m j −1 (u g + y j −1 )

Figure 2. Dynamic equilibrium of free body.

Copyright © 2012 John Wiley & Sons, Ltd. Struct. Control Health Monit. (2012)
DOI: 10.1002/stc
M. KUWABARA, S. YOSHITOMI AND I. TAKEWAKI

X
N   
ksj vj þ csj v_ j ¼ mi u€ g þ €y i (4a)
i¼j
N n
X   o XN  
kbj fj þ cbj f_ j ¼ mi u€ g þ €y i Hit  Hj1
t
þ Ji €θ i ; (4b)
i¼j i¼j
P
whereHit ¼ ij¼1 Hj. It will be assumed later that the second term in the right-hand side of Equation (4b) can
be neglected compared with the first term. Fourier transformation of Equations (3a), (3b), (4a), and (4b) and
arrangement of the resulting equations lead to
Vj ¼ Yj  Yj1  Yj Hj (5a)

X
j
Yj ¼ Φm (5b)
m¼1
  X
N   
ksj þ iocsj Vj ¼ mi U€ g þ Y€ i (6a)
i¼j
  N n
X   o
kbj þiocbj Φj ¼ mi U€ g þ Y€ i Hit  Hj1
t
(6b)
i¼j

In Equations (5a) (5b), (6a), and (6b), Vj, Yj, Yj, Φj, Üg are Fourier transforms of vj, yj, θj, fj, üg,
respectively.
Substitution of Equations (5b) and (6b) into Equation (5a) provides
Xj
Vj ¼ Yj  Yj1  Hj Φm
m¼1 " # (7)
Xj
1 XN    t  
¼ Yj  Yj1  Hj € €
mi U g þ Y i Hi  Hm1 t

m¼1 bm
k þ iocbm i¼m

Further substitution of Equation (7) into Equation (6a) and some manipulation yield
" #
     X j
1 XN    t 
ksj þ iocsj Yj  Yj1  ksj þ iocsj Hj mi U€ g þ Y€ i Hi  Hm1
t

m¼1 bm
k þ iocbm i¼m
XN   
¼ mi U € g þ Y€ i
i¼j

(8)
Application of the relation Ÿ = (i o)2Y to Equation (8) and further manipulation of the resulting
equation lead to the following expression in terms of only the floor accelerations {Üg + Ÿi}.
PN    2 PN   t   3
o2 mi U€ g þ Y€ i m H  H t € g þ Y€ i
U
1 X6
j
1
i i m1 7

i¼j
   o2 Hj 6 i¼m
   7
€ € € €
ksj þ iocsj U g þ Y j  U g þ Y j1 4 kbm þ iocbm € € € €
U g þ Y j  U g þ Y j1 5¼1
m¼1

(9)
3. NEW IDENTIFICATION ALGORITHM
Let us introduce the following function in analogy from shear building models [25]:
0 1
B o2 Mj C Mj
fj ðoÞ ¼ B
@

A 2P 3; (10)
€ g þY€ j1
U N

€ g þY€ j  1
U Mj Pj 6 f ð mi Hit Hm1
t
Þ g7
þ Hj 6 i¼m 7
ksj þiocsj 4 kbm þiocbm 5
m¼1

Copyright © 2012 John Wiley & Sons, Ltd. Struct. Control Health Monit. (2012)
DOI: 10.1002/stc
SYSTEM IDENTIFICATION AND DAMAGE DETECTION OF HIGH-RISE BUILDINGS

P
where Mj ¼ Ni¼j mi . Equation (10) is the definition of the function fj(o) and has been derived from
Equation (9) by assuming Üg + Ÿi = Üg + Ÿj(for all i). This assumption can be validated at o ! 0, that
is, Üg + Ÿi ! Üg for o ! 0. Because Equation (10) is used for o ! 0 in the stiffness and damping
identification, this assumption does not cause any problem. The introduction of Equation (10) plays
a central and key role in the present formulation. It should be noted that only the application of
dynamic equilibrium, Equation (9), does not lead to the identification function of Equation (10) of
simple form including only two adjacent acceleration terms Üg + Ÿj1 and Üg + Ÿj. Equation (10) has
been introduced in comparison with the corresponding function for shear building models [25]. In
other words, when an assumption Üg + Ÿi =Üg + Ÿj(for all i) is introduced in Equation (9) and the same
form as that for shear building models is used as the identification function, a simple form including
only two adjacent acceleration terms Üg + Ÿj1 and Üg + Ÿj appropriate for identification of shear stiff-
ness (jth story) and bending stiffness (first to jth story) can be derived.
Finally, the following formulas are derived: 2PN    3
m H t
 H t
      1 Hj X 6 7
j i i m1
lim Re 1=fj ðoÞ ¼ lim Re Fj ðoÞ ¼ þ 6i¼m 7 (11a)
o!0 o!0 ksj Mj 4 kbm 5
m¼1

PN  2  3

c m H t
 H t
d    Hj X 6 7
j bm i i m1
d 1 csj 6 i¼m 7
lim Im ¼ lim Im Fj ðoÞ ¼  2 
o!0 do f j ðo Þ o!0 do ksj Mj m¼1 4 kbm 2 5

(11b)
In Equations (11a) and (11b), Fj(o)  1/fj(o). Equation (11a) indicates that the terms including the
extensional and rotational springs of the shear-bending model can be obtained as a limit value of the
function including the transfer function with respect to two consecutive stories. On the other hand,
Equation (11b) means that the terms including the extensional and rotational dashpots of the
shear-bending model for identified stiffnesses can be obtained as a limit value of the derivative of
the function including the transfer function with respect to two consecutive stories. It should be noted
that both equations constitute recursive equations.
When the stiffnesses of the extensional and rotational springs are desired to be identified, their ratios
have to be specified. This requirement also exists in the identification of damping coefficients of
dashpots. However, the specification of the ratios between the extensional and rotational spring
stiffnesses is not necessary in the damage detection procedure explained later. The application of
Equation (11a) to damage detection will be discussed in Section 5.

4. IDENTIFICATION OF STIFFNESS AND DAMPING COEFFICIENT BY USING ARX MODEL


In the previous method [25,26] for shear building models, the limit manipulation of the identification
function fj(o) for o ! 0 is used. However, it is often the case that the identification function becomes
unstable and exhibits a large variability in the low-frequency range. To overcome this difficulty, an
ARX model is introduced, which is a time-domain model. The reliability of the ARX model in this
direction has been confirmed in [31,32]. Especially in [32], the applicability of the ARX model to shear
building models has been demonstrated.
4.1. Taylor series expansion of transfer function
Rewrite Equation (10) as follows.
o2 M j
fj ðoÞ ¼  1
; (12)
Gj;j1 ðo Þ  1
where Gj,j1(o) is the transfer function between jth and (j  1)th floors defined by
€ g þ Y€ j
U
Gj;j1 ðoÞ ¼ (13)
€ g þ Y€ j1
U
This transfer function can also be expressed in terms of the ARX parameters (Appendix A).

Copyright © 2012 John Wiley & Sons, Ltd. Struct. Control Health Monit. (2012)
DOI: 10.1002/stc
M. KUWABARA, S. YOSHITOMI AND I. TAKEWAKI

b1 eioT0 þ ⋯ þ bn einoT0
Gj;j1 ðoÞ ¼ (14)
1 þ a1 eioT0 þ ⋯ þ an einoT0
Consider the Taylor series expansion of Gj, j1(o) as follows.
Gj;j1 ðoÞ ’ A0 þ A1 o þ A2 o2 þ ⋯ (15)
From Equations (14) and (15), the coefficients A0, A1, A2 of the Taylor series expansion can be
expressed in terms of the ARX parameters {ak}, {bk}.
P
n
bk
A0 ¼  k¼1
 (16)
P
n
1þ ak
k¼1

nP



1 P
n P
n nP
1
ðn  k Þbk 1 þ ak  bk n þ ðn  k Þak
A1 ¼ iT0 k¼1 k¼1

k¼1
2
k¼1
(17)
P
n
1þ ak

½
k¼1
( ) ( )
X
n1 X
n X
n1
2
X
n1 X
n1
ðn  kÞ2 bk bk n2 þ ðn  k Þ ak ðn  k Þbk n þ ðn  k Þak
T0 2 k¼1 k¼1 k¼1 k¼1 k¼1
A2 ¼  X
n  ( )2 2 ( )2
2 X
n X
n
1þ ak 1þ ak 1þ ak


k¼1
k¼1 k¼1
( )2
X
n X
n1
bk n þ ðn  k Þak
þ2
k¼1 k¼1 (18)
( )3
X
n
1þ ak
k¼1

By introducing the real and imaginary parts of Aj as


Aj ¼ ARj þ iAIj (19)
and substituting the properties of the real and imaginary parts of Aj, the transfer function can be reduced to
   
Gj;j1 ðoÞ ’ AR0 þ AR2 o2 þ ⋯ þ i AI1 o þ AI3 o3 þ ⋯ (20)

4.2. Relation of the limit value of transfer function with ARX parameters
It is meaningful to note that Equation (21) can be derived from the mechanical interpretation, that is,
the jth floor and (j  1)th floor move identically at o ! 0.
 
lim Re Gj;j1 ðoÞ ¼ 1 (21)
o!0
The following relation also holds because Gj, j1(o) should not include linear terms of o judging
from Equations (11a), (11b), and (12).
d  
lim Im Gj;j1 ðoÞ ¼ 0 (22)
o!0 do

Equation (21) leads to Equation (23), and Equation (22) yields Equation (24).
AR0 ¼ 1 (23)

AI1 ¼ 0 (24)
Furthermore, the following equations can be derived from Equations (16), (17), (23), and (24).

Copyright © 2012 John Wiley & Sons, Ltd. Struct. Control Health Monit. (2012)
DOI: 10.1002/stc
SYSTEM IDENTIFICATION AND DAMAGE DETECTION OF HIGH-RISE BUILDINGS

X
n X
n
ak þ 1 ¼ bk (25)
k¼1 k¼1
( ) ( )
X
n1 X
n X
n X
n1
ðn  kÞbk 1 þ ak ¼ bk n þ ðn  kÞak (26)
k¼1 k¼1 k¼1 k¼1

These relations will be used as the constraints in the estimation of the ARX parameters. Equation
(25) is used for stiffness identification and Equation (26) for damping identification.

4.3. ARX parameter estimation with constraints


The proposed method requires the prediction of the ARX parameters under the conditions on the ARX
parameters (Equation (25) for stiffness identification). It is found that Equation (25) is a linear equation
for the ARX parameters. A batch processing least-squares estimation method [33] (Appendices B and C)
provides

Rθ ¼ f (27)
By applying the Lagrange multiplier method, the linear constraint can be incorporated into the batch
processing least-squares estimation method.
Equation (25) can be expressed by

pT θ ¼ 1; (28)
where

p ¼ f1; . . . ; 1; 1; . . . ; 1gT (29)


Therefore, the present method is reduced to the problem for solving the following equation.



R p θ f
¼ (30)
pT 0 l 1
4.4. ARX parameter expression of identification function
Substitution of Equations (23) and (24) into Equation (20) leads to
   
Gj;j1 ðoÞ ’ 1 þ AR2 o2 þ ⋯ þ i AI3 o3 þ ⋯ (31)

Substitution of Equation (31) into Equation (12) and transformation of the denominator into the real
number provide
    2
1 þ AR2 o2 þ ⋯ AR2 o2 þ ⋯ þ AI3 o3 þ⋯
f j ð o Þ ¼ o Mj
2
 R   
A o 2 þ ⋯ 2 þ AI o3 þ⋯ 2
 R 2 2   
3   (32)
A o þ ⋯ AI3 o3 þ⋯  1 þ AR2 o2 þ ⋯ AI3 o3 þ⋯
þio2 Mj 2  R 2  2
A2 o2 þ ⋯ þ AI3 o3 þ⋯

From Equation (32), the stiffness can be identified as


2PN   3
X mi Hit  Hm1
t
      1 Hj
j
6i¼m 7 AR2
lim Re 1=fj ðoÞ ¼ lim Re Fj ðoÞ ¼ þ 6 7¼ (33)
o!0 o!0 ksj Mj m¼1 4 kbm 5 Mj

Because AR2 is expressed in terms of the ARX parameters as shown in Equation (18), the stiffness
identification problem can be reduced to the estimation problem of the ARX parameters. This means
that the proposed method enables one to avoid the setting of approximation functions, which is
sometimes difficult in the previous method [25].

Copyright © 2012 John Wiley & Sons, Ltd. Struct. Control Health Monit. (2012)
DOI: 10.1002/stc
M. KUWABARA, S. YOSHITOMI AND I. TAKEWAKI

5. APPLICATION OF PROPOSED IDENTIFICATION ALGORITHM TO DAMAGE DETECTION


It is shown here that the function Re[Fj(o)] in Equations (11a) and (33) can be used for damage
detection. The procedure is very simple and compares the function Re[Fj(o)] for the models before
and after damage. It can be understood from Equation (33) that, if the shear stiffness decreases in a
damaged story, the function Re[Fj(o)], more exactly its limit value at o ! 0, for that story increases.
For example, if a beam is damaged, it is expected that the story shear stiffnesses at the stories adjacent
to the beam will decrease. On the other hand, if a column is damaged, it is expected that the story shear
stiffness at the corresponding story will be reduced. The outline of the proposed damage detection
method is shown in Figure 3. The quantitative investigation on the accuracy and reliability of the
proposed method will be presented in the following section.

6. NUMERICAL EXAMPLES
6.1. Application of proposed damage detection method
6.1.1. Frame properties, input ground motions, and damage patterns. Consider a 20-story three-span
frame as shown in Figure 4. The story height is 5 m in the first and second stories and 4 m in the 3rd to
20th stories. The common span length is 7 m. The horizontal floor mass is 0.12  106 kg in all the
stories. The nodal masses in the vertical and rotational directions have also been considered. The
member properties are shown in Table I. Young’s modulus is assumed to be 205,000 N/mm2, and
the member shear deformation has been neglected. All the horizontal nodal displacements in the
same floor are assumed to be the same, and the axial deformation of beams is neglected. It is
possible to construct a shear-bending building model for this frame. However, the proposed damage
detection method does not require the construction of the shear-bending model.
A response analysis of the frame is conducted for El Centro NS 1940, Taft EW 1952, and
Hachinohe NS 1968, and the response result is used as substitute of recorded data. The result for El
Centro NS 1940 is shown first. The number (na = nb) of order of the ARX model has been selected
as 50–70 after several convergence investigations. Investigation on the determination of the number
of order of the ARX model will be carried out in Section 6.4.
Thirteen patterns are dealt with as beam damage patterns and five patterns as column damage
patterns. Those damage patterns are shown in Figures 4–8. For essential presentation of properties
of the proposed method, only the results for eight patterns of beam damage (patterns 1–8) and two

Unchanged
no damage
u j (t) Absolute horizontal
u j −1 (t) acceleration Re {F j ( )} After damage
Before damage

Transfer function by ARX model

U j( ) b1e − i T0 + + bne −n i T0 increase damage


Gj ( ) = =
U j −1 ( ) 1+ a1e − i T0 + +ane −n i T0 After damage

Identification function of
Re {F j ( )}
shear-bending model
Before damage
N −1

− 2
mi
u g (t)
Fj ( )= i= j Comparison of identification functions
1 at =0 before and after damage
−1
Gj ( )

Figure 3. Outline of proposed damage detection method.

Copyright © 2012 John Wiley & Sons, Ltd. Struct. Control Health Monit. (2012)
DOI: 10.1002/stc
SYSTEM IDENTIFICATION AND DAMAGE DETECTION OF HIGH-RISE BUILDINGS

Pattern 5 50% decrease in


20
beam stiffness
19
18
17 Pattern 4 50% decrease in
16
beam stiffness
15
14
13
12 Pattern 3 50% decrease in
11
beam stiffness
10
9
8
7 Pattern 2 50% decrease in
6
beam stiffness
5
4
3
Pattern 1 50% decrease in
2
beam stiffness
1

Figure 4. Damage patterns 1–5 (damage in beam).

Table I. Member properties.


Column Beam
2 4
Story Cross-sectional area (m ) Second moment of area (m ) Second moment of area (m4)
20 0.0404 0.00190 0.00172
19 0.0404 0.00190 0.00172
18 0.0404 0.00190 0.00172
17 0.0404 0.00190 0.00172
16 0.0404 0.00190 0.00172
15 0.0465 0.00216 0.00292
14 0.0465 0.00216 0.00292
13 0.0465 0.00216 0.00292
12 0.0465 0.00216 0.00292
11 0.0465 0.00216 0.00292
10 0.0553 0.00364 0.00292
9 0.0553 0.00364 0.00292
8 0.0553 0.00364 0.00292
7 0.0553 0.00364 0.00292
6 0.0553 0.00364 0.00292
5 0.0809 0.00704 0.00411
4 0.0809 0.00704 0.00411
3 0.0809 0.00704 0.00411
2 0.0809 0.00704 0.00411
1 0.0809 0.00704 0.00411

patterns of column damage (patterns 14 and 16) are shown in this paper. Table II shows the change of
the fundamental natural period of the frame before and after damage. It can be observed that the differ-
ence is quite small. This is because the fundamental natural period is a global performance parameter of
the frame and local damage is difficult to be reflected in this global performance parameter. Figure 9
illustrates the interstory drifts in the lowest eigenmode multiplied by the lowest-mode participation
factor of the models before and after damage. It can be observed that a clear difference is seen in
patterns 1–4, 8, 14, and 16. It should be noted that the identification of all the floor displacements in
the lowest eigenmode requires data measurements in multiple floors, and this is quite cumbersome.
The 50% damage is the beam stiffness reduction, not the story stiffness reduction. In Northridge and
Kobe earthquakes, we experienced beam-end fractures under severe shaking. If a member end causes
fracture and is modeled by a pin joint, this level of damage may be possible.

Copyright © 2012 John Wiley & Sons, Ltd. Struct. Control Health Monit. (2012)
DOI: 10.1002/stc
M. KUWABARA, S. YOSHITOMI AND I. TAKEWAKI

20
19
18
17
16
15
14
13
12
11
10
9
8
Pattern 6 7
6
50% decrease 5
4
in beam stiffness 3
2
1

20 20
19 19
18 18
17 17
16 16
15 Pattern 7 15
14 14
13 50% decrease 13
12 12
11
in beam stiffness 11
10 10
9 9
8 8
7 7
6
5
Pattern 8 6
5
4 50% decrease 4
3 3
2
in beam stiffness 2
1 1

Figure 5. Damage patterns 6–8 (partial damage in beam).

20 20
19 19
18 Pattern 9 18
17 50% decrease 17
16 16
15 in beam stiffness 15
14 14
13 13
12 12
11 11
10 10
9 9
8 8
7 Pattern 10 7
6 6
5 50% decrease 5
4 4
3 in beam stiffness 3
2 2
1 1

20 20
19 19
18
Pattern 11 18
17 50% decrease 17
16 16
15 in beam stiffness 15
14 14
13 13
12 12
11 11
10 10
9 9
8 8
7 Pattern 12 7
6 6
5 50% decrease 5
4 4
3 in beam stiffness 3
2 2
1 1

Figure 6. Damage patterns 9–12 (damage in two-floor beams).

6.1.2. Damage patterns in beams. Figure 10 shows the comparison of the function defined in Equation
(33) for models before and after beam damage (damage pattern 1; second-story floor beam). It can be
observed that the identification functions before and after damage for the first and second stories

Copyright © 2012 John Wiley & Sons, Ltd. Struct. Control Health Monit. (2012)
DOI: 10.1002/stc
SYSTEM IDENTIFICATION AND DAMAGE DETECTION OF HIGH-RISE BUILDINGS

20
19
18
17
16
15 Pattern 13
14
13
50% decrease
12 in beam stiffness
11
10
9
8
7
6
5
4
3
2
1

Figure 7. Damage pattern 13 (damage in multifloor beams).

Pattern 18
20
19
50% decrease in column stiffness
18
17
16 Pattern 17
15 50% decrease in column stiffness
14
13
12
11 Pattern 16
10 50% decrease in column stiffness
9
8
7
6 Pattern 15
5 50% decrease in column stiffness
4
3
2 Pattern 14
1
50% decrease in column stiffness

Figure 8. Damage patterns 14–18 (damage in columns).

Table II. Change of fundamental natural period before and after damage.
Damage pattern 1 2 3 4 5 6 7 8 14 16
Before damage (s) 2.195
After damage (s) 2.224 2.219 2.218 2.205 2.195 2.202 2.201 2.207 2.228 2.215
Change (%) +1.32 +1.09 +1.05 +0.46 0.00 +0.32 +0.27 +0.55 +1.50 +0.91

exhibit remarkable differences. Although only the identification functions before and after damage for
the third story are presented, those for other stories above the third story exhibit a property similar to
the third story.
The identification function, Equation (10), at zero frequency has been derived from the dynamic
equilibrium of the model as shown in Section 2.2. The identification function except at zero frequency
has not been derived from the dynamic equilibrium. However, only the value at zero frequency is used
for damage detection. The identification function except at zero frequency has been used only for
comparison with the expression for shear building models. In addition, the presentation of the
identification function except at zero frequency helps the confirmation of computational stability of this
function around zero frequency. Only the presentation of the function value at zero frequency cannot
guarantee such computational stability.
Figures 11–14 present the comparison of the function defined in Equation (33) for models before
and after beam damage (damage patterns 2–5; 6th-story floor beam, 11th-story floor beam, 16th-story
floor beam, 20th-story upper beam). It can be understood from Figure 11 that the identification
functions before and after damage for the fifth and sixth stories exhibit remarkable differences and

Copyright © 2012 John Wiley & Sons, Ltd. Struct. Control Health Monit. (2012)
DOI: 10.1002/stc
M. KUWABARA, S. YOSHITOMI AND I. TAKEWAKI

pattern 1 pattern 2 pattern 3


20 20 20

15 15 15
before damage before damage before damage
after damage after damage after damage
story

story

story
10 10 10

5 5 5

0 0 0
0 0 .02 0 .04 0 .06 0 .08 0 .1 0 .12 0 0 .02 0 .04 0 .06 0 .08 0 .1 0 .12 0 0 .02 0 .04 0 .06 0 .08 0 .1 0 .12
participation factor × eigenmode participation factor × eigenmode participation factor × eigenmode

pattern 4 pattern 5 pattern 6


20 20 20

15 15 15
before damage before damage before damage
after damage after damage after damage
story

story

story
10 10 10

5 5 5

0 0 0
0 0 .02 0 .04 0 .06 0 .08 0 .1 0 .12 0 0 .02 0 .04 0 .06 0 .08 0 .1 0 .12 0 0 .02 0 .04 0 .06 0 .08 0 .1 0 .12
participation factor × eigenmode participation factor × eigenmode participation factor × eigenmode

pattern 7 pattern 8 pattern 14


20 20 20

15 15 15
before damage before damage before damage
after damage after damage after damage
story

story

story

10 10 10

5 5 5

0 0 0
0 0 .02 0 .04 0 .06 0 .08 0 .1 0 .12 0 0 .02 0 .04 0 .06 0 .08 0 .1 0 .12 0 0 .02 0 .04 0 .06 0 .08 0 .1 0 .12
participation factor × eigenmode participation factor × eigenmode participation factor × eigenmode

pattern 16
20

15
before damage
after damage
story

10

0
0 0 .02 0 .04 0 .06 0 .08 0 .1 0 .12
participation factor × eigenmode

Figure 9. Interstory drifts in the lowest eigenmode multiplied by the lowest-mode participation factor of the
models before and after damage (patterns 1–8, 14, and 16).

those for the fourth and seventh stories do not. This corresponds well with the result for Figure 10.
Similar tendencies can be observed from Figures 12–14.
Figure 15 shows the comparison of the function defined in Equation (33) for models before and after
beam damage (damage pattern 6; only mid-span beam in the 11th-story floor). It can be understood that
the difference of the identification functions before and after damage is not remarkable even in the 10th
and 11th stories. This is because the influence of the damage of the mid-span beam on the story

Copyright © 2012 John Wiley & Sons, Ltd. Struct. Control Health Monit. (2012)
DOI: 10.1002/stc
SYSTEM IDENTIFICATION AND DAMAGE DETECTION OF HIGH-RISE BUILDINGS

1st story 2nd story


2 10-8 2 10-8

Identification Function Re[F( )]

Identification Function Re[F( )]


before damage before damage
after damage after damage
-8 -8
1.5 10 1.5 10

1 10-8 1 10-8

5 10-9 5 10-9

0 0
0 0.5 1 1.5 2 0 0.5 1 1.5 2
circular frequency [rad/s] circular frequency [rad/s]

3rd story
2 10-8
Identification Function Re[F( )]

before damage
after damage
1.5 10-8

1 10-8

5 10-9

0
0 0.5 1 1.5 2
circular frequency [rad/s]

Figure 10. Identification functions before and after damage for pattern 1 (damage in beams): El Centro
NS 1940.

4th story 5th story


2 10-8 2 10-8
Identification Function Re[F( )]

Identification Function Re[F( )]

before damage before damage


after damage after damage
1.5 10-8 1.5 10-8

1 10-8 1 10-8

5 10-9 5 10-9

0 0
0 0.5 1 1.5 2 0 0.5 1 1.5 2
circular frequency [rad/s] circular frequency [rad/s]

6th story 7th story


2 10-8 2 10-8
Identification Function Re[F( )]

Identification Function Re[F( )]

before damage before damage


after damage after damage
-8 -8
1.5 10 1.5 10

1 10-8 1 10-8

-9 -9
5 10 5 10

0 0
0 0.5 1 1.5 2 0 0.5 1 1.5 2
circular frequency [rad/s] circular frequency [rad/s]

Figure 11. Identification functions before and after damage for pattern 2 (damage in beams).

Copyright © 2012 John Wiley & Sons, Ltd. Struct. Control Health Monit. (2012)
DOI: 10.1002/stc
M. KUWABARA, S. YOSHITOMI AND I. TAKEWAKI

9th story 10th story


2 10-8 2 10-8

Identification Function Re[F( )]

Identification Function Re[F( )]


before damage before damage
after damage after damage
1.5 10-8 1.5 10-8

1 10-8 1 10-8

5 10-9 5 10-9

0 0
0 0.5 1 1.5 2 0 0.5 1 1.5 2
circular frequency [rad/s] circular frequency [rad/s]

11th story 12th story


2 10-8 2 10-8

Identification Function Re[F( )]


Identification Function Re[F( )]

before damage before damage


after damage after damage
1.5 10-8 1.5 10-8

1 10-8 1 10-8

5 10-9 5 10-9

0 0
0 0.5 1 1.5 2 0 0.5 1 1.5 2
circular frequency [rad/s] circular frequency [rad/s]

Figure 12. Identification functions before and after damage for pattern 3 (damage in beams).

14th story 15th story


2 10-8 2 10-8
Identification Function Re[F( )]

Identification Function Re[F( )]

before damage before damage


after damage after damage
1.5 10-8 1.5 10-8

1 10-8 1 10-8

5 10-9 5 10-9

0 0
0 0.5 1 1.5 2 0 0.5 1 1.5 2
circular frequency [rad/s] circular frequency [rad/s]

16th story 17th story


2 10-8 2 10-8
Identification Function Re[F( )]

Identification Function Re[F( )]

1.5 10-8 1.5 10-8

1 10-8 1 10-8

before damage
after damage
5 10-9 5 10-9
before damage
after damage
0 0
0 0.5 1 1.5 2 0 0.5 1 1.5 2
circular frequency [rad/s] circular frequency [rad/s]

Figure 13. Identification functions before and after damage for pattern 4 (damage in beams).

Copyright © 2012 John Wiley & Sons, Ltd. Struct. Control Health Monit. (2012)
DOI: 10.1002/stc
SYSTEM IDENTIFICATION AND DAMAGE DETECTION OF HIGH-RISE BUILDINGS

19th story 20th story


4 10-8 4 10-8

Identification Function Re[F( )]

Identification Function Re[F( )]


before damage
3.5 10-8 after damage 3.5 10-8
-8
3 10 3 10-8
-8
2.5 10 2.5 10-8
-8
2 10 2 10-8
-8
1.5 10 1.5 10-8

1 10-8 1 10-8 before damage


-9 -9 after damage
5 10 5 10

0 0
0 0.5 1 1.5 2 0 0.5 1 1.5 2
circular frequency [rad/s] circular frequency [rad/s]

Figure 14. Identification functions before and after damage for pattern 5 (damage in beams).

9th story 10th story


2 10-8 2 10-8
Identification Function Re[F( )]

Identification Function Re[F( )]


before damage before damage
after damage after damage
1.5 10-8 1.5 10-8

1 10-8 1 10-8

5 10-9 5 10-9

0 0
0 0.5 1 1.5 2 0 0.5 1 1.5 2
circular frequency [rad/s] circular frequency [rad/s]

11th story 12th story


2 10-8 2 10-8
Identification Function Re[F( )]

Identification Function Re[F( )]

before damage before damage


after damage after damage
1.5 10-8 1.5 10-8

1 10-8 1 10-8

-9 -9
5 10 5 10

0 0
0 0.5 1 1.5 2 0 0.5 1 1.5 2
circular frequency [rad/s] circular frequency [rad/s]

Figure 15. Identification functions before and after damage for pattern 6 (damage in beam).

stiffness is not significant. Figure 16 presents the comparison of the function defined in Equation (33)
for models before and after beam damage (damage pattern 7; only left-span beam in the 11th-story
floor). It can be understood that the difference of the identification functions before and after damage
is not remarkable even in the 10th and 11th stories. This is because, whereas the influence of the
damage of the one-span beam on the story stiffness is not significant, that of the two-span beams is
slightly significant. Figure 17 indicates the comparison of the function defined in Equation (33) for
models before and after beam damage (damage pattern 8; only the side-span beams in the 11th-story
floor). It can be understood that the difference of the identification functions before and after damage
is slightly remarkable in the 10th and 11th stories compared with Figures 15 and 16. This is because the
influence of the damage of the left-span beams on the story stiffness is not significant.
The present damage detection method has a limitation that, if the fourth and sixth floors are
damaged simultaneously for example, we can notice only that some damages occur around the fourth
to seventh stories. This limitation applies to pattern 11–13.

Copyright © 2012 John Wiley & Sons, Ltd. Struct. Control Health Monit. (2012)
DOI: 10.1002/stc
M. KUWABARA, S. YOSHITOMI AND I. TAKEWAKI

9th story 10th story


2 10-8 2 10-8

Identification Function Re[F( )]

Identification Function Re[F( )]


before damage before damage
after damage after damage
-8 -8
1.5 10 1.5 10

1 10-8 1 10-8

5 10-9 5 10-9

0 0
0 0.5 1 1.5 2 0 0.5 1 1.5 2
circular frequency [rad/s] circular frequency [rad/s]

11th story 12th story


2 10-8 2 10-8
Identification Function Re[F( )]

Identification Function Re[F( )]


before damage before damage
after damage after damage
-8 -8
1.5 10 1.5 10

1 10-8 1 10-8

5 10-9 5 10-9

0 0
0 0.5 1 1.5 2 0 0.5 1 1.5 2
circular frequency [rad/s] circular frequency [rad/s]

Figure 16. Identification functions before and after damage for pattern 7 (damage in beam).

-8
9th story -8
10th story
2 10 2 10
Identification Function Re[F( )]

Identification Function Re[F( )]

before damage before damage


after damage after damage
1.5 10-8 1.5 10-8

1 10-8 1 10-8

5 10-9 5 10-9

0 0
0 0.5 1 1.5 2 0 0.5 1 1.5 2
circular frequency [rad/s] circular frequency [rad/s]

11th story 12th story


2 10-8 2 10-8
Identification Function Re[F( )]

Identification Function Re[F( )]

before damage before damage


after damage after damage
-8 -8
1.5 10 1.5 10

1 10-8 1 10-8

5 10-9 5 10-9

0 0
0 0.5 1 1.5 2 0 0.5 1 1.5 2
circular frequency [rad/s] circular frequency [rad/s]

Figure 17. Identification functions before and after damage for pattern 8 (damage in beams).
6.1.3. Damage patterns in columns. On the other hand, Figure 18 shows the comparison of the
function defined in Equation (33) for models before and after column damage (damage pattern 14;
first-story columns). It can be observed that the identification functions before and after damage for

Copyright © 2012 John Wiley & Sons, Ltd. Struct. Control Health Monit. (2012)
DOI: 10.1002/stc
SYSTEM IDENTIFICATION AND DAMAGE DETECTION OF HIGH-RISE BUILDINGS

1st story 2nd story


2 10-8 2 10-8

Identification Function Re[F( )]

Identification Function Re[F( )]


before damage before damage
after damage after damage
-8 -8
1.5 10 1.5 10

1 10-8 1 10-8

-9 -9
5 10 5 10

0 0
0 0.5 1 1.5 2 0 0.5 1 1.5 2
circular frequency [rad/s] circular frequency [rad/s]

Figure 18. Identification functions before and after damage for pattern 14 (damage in columns).

the first story exhibit remarkable differences. Although only the identification functions before and
after damage for the second story are presented, those for other stories above the second story
exhibit a property similar to the second story.
Figure 19 presents the comparison of the function defined in Equation (33) for models before
and after column damage (damage pattern 16; 10th-story columns). It can be understood that the
identification functions before and after damage for the 10th story exhibit remarkable differences.

6.2. Accuracy and reliability of proposed damage detection procedure


To verify the accuracy and reliability of the procedure of using Equation (33) including the transfer
function constructed by the ARX model, Equation (33) that is derived from the exact transfer function,
which is computed from the equations of motion in frequency domain, has been evaluated. Figure 20(a)
shows the identification function before and after damage for the damage pattern 3 derived from the exact
transfer function. It can be observed that the identification function, Figure 12, evaluated using the

9th story 10th story


2 10-8 2 10-8
Identification Function Re[F( )]

Identification Function Re[F( )]

before damage before damage


after damage after damage
-8 -8
1.5 10 1.5 10

1 10-8 1 10-8

5 10-9 5 10-9

0 0
0 0.5 1 1.5 2 0 0.5 1 1.5 2
circular frequency [rad/s] circular frequency [rad/s]

11th story
2 10-8
Identification Function Re[F( )]

before damage
after damage
-8
1.5 10

1 10-8

5 10-9

0
0 0.5 1 1.5 2
circular frequency [rad/s]

Figure 19. Identification functions before and after damage for pattern 16 (damage in columns).

Copyright © 2012 John Wiley & Sons, Ltd. Struct. Control Health Monit. (2012)
DOI: 10.1002/stc
M. KUWABARA, S. YOSHITOMI AND I. TAKEWAKI

(a) 9th story 10th story


2 10-8 2 10-8

Identification Function Re[F( )]

Identification Function Re[F( )]


before damage before damage
after damage after damage
-8 -8
1.5 10 1.5 10

1 10-8 1 10-8

5 10-9 5 10-9

0 0
0 0 .5 1 1 .5 2 0 0 .5 1 1 .5 2
circular frequency [rad/s] circular frequency [rad/s]

11th story 12th story


2 10-8 2 10-8
Identification Function Re[F( )]

Identification Function Re[F( )]


before damage before damage
after damage after damage
-8 -8
1.5 10 1.5 10

1 10
-8
1 10-8

5 10-9 5 10-9

0 0
0 0 .5 1 1 .5 2 0 0.5 1 1.5 2
circular frequency [rad/s] circular frequency [rad/s]

(b) 9th story 10th story


2 10-8 2 10-8
Identification Function Re[F( )]

Identification Function Re[F( )]

before damage before damage


after damage after damage
-8
1.5 10 1.5 10 -8

1 10-8 1 10-8

5 10-9 5 10
-9

0 0
0 0 .5 1 1 .5 2 0 0 .5 1 1 .5 2
circular frequency [rad/s] circular frequency [rad/s]

11th story
2 10-8
Identification Function Re[F( )]

before damage
after damage
-8
1.5 10

1 10-8

5 10-9

0
0 0 .5 1 1 .5 2
circular frequency [rad/s]

Figure 20. (a) Identification functions before and after damage for pattern 3 derived from exact transfer function
(damage in beams); (b) Identification functions before and after damage for pattern 16 derived from exact transfer
function (damage in columns).

ARX model for El Centro NS 1940 coincides fairly well with that, Figure 20(a), derived from the
exact transfer function. Figure 20(b) presents that for the damage pattern 16. A similar observation
can be seen.

Copyright © 2012 John Wiley & Sons, Ltd. Struct. Control Health Monit. (2012)
DOI: 10.1002/stc
SYSTEM IDENTIFICATION AND DAMAGE DETECTION OF HIGH-RISE BUILDINGS

The quantitative analysis of the increased ratio of the identification function at zero frequency
(theoretical; transfer function has been evaluated from the equations of motion) is shown in Table III
and that evaluated for the ground motion of El Centro NS 1940 is presented in Table IV. The data for
damage patterns 9 and 12 (damage in multiple stories) are also shown in Tables III and IV. It can be
observed that the increased ratio of the identification function at zero frequency evaluated for El Centro
NS 1940 is reliable as a measure for damage detection.
The bending stiffness of a shear-bending model is related primarily with the axial stiffness of
columns. But only the case of damage in the shear stiffness has been dealt with in the aforementioned

Table III. Increased ratio of identification function at zero frequency (theoretical; %).
Pattern (damage in beam) Pattern (damage in column)
Story 1 2 3 4 5 6 7 8 9 12 14 16
20 0.00 +2.98
19 0.00 +0.35
18
17 +1.05 +1.04
16 +9.79 +9.79
15 +9.61 +9.61
14 +0.68 +0.69
13 +1.73
12 +0.97 +0.55 +0.11 +0.23 +15.0
11 +12.7 +3.29 +3.40 +7.19 +26.1 +0.44
10 +14.9 +3.80 +3.96 +8.39 +16.4 +27.2
9 +1.54 +0.66 +0.28 +0.59 +1.73 +0.44
8
7 +1.46 +1.47
6 +14.2 +14.2
5 +18.8 +18.8
4 +2.55 +2.55
3 +2.53
2 +19.3 +2.43
1 +24.9 +58.8

Table IV. Increased ratio of identification function at zero frequency (El Centro NS 1940; %).
Pattern (damage in beam) Pattern (damage in column)
Story 1 2 3 4 5 6 7 8 9 12 14 16
20 0.01 +2.97
19 0.01 +0.39
18
17 +0.98 2.20
16 +9.84 +9.70
15 +14.7 +14.2
14 0.46 4.18
13 +2.02
12 1.49 +0.47 +0.89 0.60 +12.9
11 +11.8 +3.42 +3.28 +6.55 +31.0 +0.45
10 +20.6 +1.76 +2.80 +8.43 +17.3 +29.4
9 0.97 +2.46 +1.44 1.90 +2.31 2.35
8
7 +1.48 +1.20
6 +14.1 +14.3
5 +18.8 +18.3
4 +2.56 +3.80
3 +2.54
2 +24.4 2.80
1 +26.0 +59.6

Copyright © 2012 John Wiley & Sons, Ltd. Struct. Control Health Monit. (2012)
DOI: 10.1002/stc
M. KUWABARA, S. YOSHITOMI AND I. TAKEWAKI

numerical examples. In this case of damage in the shear stiffness, the damage in the jth story does not
influence the identification function in the upper stories so much. An example is shown for damage
pattern 1 in Tables III and IV. The influence in the 19th and 20th stories is negligible.

6.3. Influence of noise in data


Figure 21 shows the identification function before and after damage for damage pattern 3 derived from
data with noise (20% of RMS (Root Mean Square) for the signal; white noise in 0–150 rad/s) and the
exact transfer function. Table V presents the quantitative comparison between the results without and
with noise. It can be observed that, even if noise is included to some extent, the damage detection is
possible with the proposed procedure.

6.4. Investigation for other ground motions


Figure 22 illustrates the comparison of the function defined in Equation (33) for models before and
after beam damage (damage pattern 3) and evaluated using the ARX model for Taft EW 1952 and
Figure 23 shows that for Hachinohe NS 1968. It can be observed that the function defined in Equation (33)
for models before and after damage and evaluated using the ARX model exhibits almost the same function

9th story 10th story


2 10-8 2 10-8
Identification Function Re[F( )]

Identification Function Re[F( )]


before damage before damage
after damage after damage
1.5 10-8 theoretical curve (B.D.) 1.5 10-8 theoretical curve (B.D.)
theoretical curve (A.D.) theoretical curve (A.D.)

1 10-8 1 10-8

5 10-9 5 10-9

0 0
0 0.5 1 1.5 2 0 0.5 1 1.5 2
circular frequency [rad/s] circular frequency [rad/s]

11th story 12th story


2 10-8 2 10-8
Identification Function Re[F( )]

Identification Function Re[F( )]

before damage before damage


after damage after damage
1.5 10 -8 theoretical curve (B.D.) 1.5 10 -8 theoretical curve (B.D.)
theoretical curve (A.D.) theoretical curve (A.D.)

1 10-8 1 10-8

5 10-9 5 10-9

0 0
0 0.5 1 1.5 2 0 0.5 1 1.5 2
circular frequency [rad/s] circular frequency [rad/s]

Figure 21. Identification functions before and after damage for pattern 3 derived from data with noise (20% of
RMS) and exact transfer function.

Table V. Increased ratio of identification function at zero frequency (pattern 3; damage in beam).
Pattern (damage in beam)
Story Theoretical El Centro Taft Hachinohe El Centro with 20% noise
12 +0.97 1.49 +0.97 +1.00 +2.05
11 +12.7 +11.8 +12.0 +12.0 +13.2
10 +14.9 +20.6 +15.2 +15.1 +33.5
9 +1.54 0.97 +1.91 +2.11 +1.77

Copyright © 2012 John Wiley & Sons, Ltd. Struct. Control Health Monit. (2012)
DOI: 10.1002/stc
SYSTEM IDENTIFICATION AND DAMAGE DETECTION OF HIGH-RISE BUILDINGS

9th story 10th story


2 10-8 2 10-8

Identification Function Re[F( )]

Identification Function Re[F( )]


before damage before damage
after damage after damage
-8 -8
1.5 10 1.5 10

1 10-8 1 10-8

-9
5 10 5 10-9

0 0
0 0.5 1 1.5 2 0 0.5 1 1.5 2
circular frequency [rad/s] circular frequency [rad/s]

11th story 12th story


2 10-8 2 10-8
Identification Function Re[F( )]

Identification Function Re[F( )]


before damage before damage
after damage after damage
-8 -8
1.5 10 1.5 10

1 10-8 1 10-8

5 10-9 5 10-9

0 0
0 0.5 1 1.5 2 0 0.5 1 1.5 2
circular frequency [rad/s] circular frequency [rad/s]

Figure 22. Identification functions before and after damage for pattern 3 (damage in beams) for Taft EW
1952.

9th story 10th story


2 10-8 2 10-8
Identification Function Re[F( )]

Identification Function Re[F( )]

before damage before damage


after damage after damage
1.5 10-8 1.5 10-8

1 10-8 1 10-8

5 10-9 5 10-9

0 0
0 0.5 1 1.5 2 0 0.5 1 1.5 2
circular frequency [rad/s] circular frequency [rad/s]

11th story 12th story


2 10-8 2 10-8
Identification Function Re[F( )]

Identification Function Re[F( )]

before damage before damage


after damage after damage
-8 -8
1.5 10 1.5 10

1 10-8 1 10-8

5 10-9 5 10-9

0 0
0 0.5 1 1.5 2 0 0.5 1 1.5 2
circular frequency [rad/s] circular frequency [rad/s]

Figure 23. Identification functions before and after damage for pattern 3 (damage in beams) for Hachinohe NS
1968.

Copyright © 2012 John Wiley & Sons, Ltd. Struct. Control Health Monit. (2012)
DOI: 10.1002/stc
M. KUWABARA, S. YOSHITOMI AND I. TAKEWAKI

9th story 10th story


100 100

limit value of Re[F(ω)] limit value of Re[F(ω)]

Limit Value of Identification

Limit Value of Identification


99.8 99.8

Function Re[F(ω)]

Function Re[F(ω)]
99.6 99.6

F it [% ]
Fit [%]
99.4 99.4

99.2 99.2
F it F it

0 99 0 99
0 50 100 150 200 0 50 100 150 200
number of order of ARX model number of order of ARX model
11th story 12th story
100 100
limit value of Re[F(ω)]
limit value of Re[F(ω)]
Limit Value of Identification

Limit Value of Identification


99.8 99.8
Function Re[F(ω)]

Function Re[F(ω)]
99.6 99.6

F it [% ]
Fit [%]
99.4 99.4

99.2 99.2
F it F it

0 99 0 99
0 50 100 150 200 0 50 100 150 200
number of order of ARX model number of order of ARX model
(a) Before damage

9th story 10th story


100 100

limit value of Re[F(ω)] limit value of Re[F(ω)]


Limit Value of Identification

Limit Value of Identification

99.8 99.8
Function Re[F(ω)]

Function Re[F(ω)]

99.6 99.6

F it [% ]
Fit [%]

99.4 99.4

99.2 99.2
F it F it

0 99 0 99
0 50 100 150 200 0 50 100 150 200
number of order of ARX model number of order of ARX model

11th story 12th story


100 100
limit value of Re[F(ω)]
limit value of Re[F(ω)]
Limit Value of Identification

Limit Value of Identification

99.8 99.8
Function Re[F(ω)]

Function Re[F(ω)]

99.6 99.6
F it [% ]
Fit [%]

99.4 99.4

99.2 99.2
F it F it

0 99 0 99
0 50 100 150 200 0 50 100 150 200
number of order of ARX model number of order of ARX model
(b) After damage
Figure 24. Relation of convergence of function ‘Fit’ with limit value of identification function for various numbers
of order of ARX model (damage pattern 3, El Centro NS 1940).

Copyright © 2012 John Wiley & Sons, Ltd. Struct. Control Health Monit. (2012)
DOI: 10.1002/stc
SYSTEM IDENTIFICATION AND DAMAGE DETECTION OF HIGH-RISE BUILDINGS

and the present damage detection method is effective for other ground motions. The quantitative analysis re-
sult can also be found in Table V.
It may be concluded that the proposed method for damage detection is effective when the influence
of the damage of members on the story stiffness is relatively significant.

6.5. Determination of number of order of ARX model


It is important to investigate how to determine the number of order of the ARX model. Let us introduce
the following function called ‘Fit’ [30].
0 sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi1
PNd
B ½^y ðkÞ  yðk Þ2 C
B C
B k¼1 C
Fit ¼ B1  sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi C  100½%; (34)
B PNd C
@ ½yðkÞ  y 2 A
k¼1

where Nd denotes the number of data, y(k) the recorded output data, ^y ðkÞ the output data computed from
^y ðk Þ ¼ θT wðk Þ (Equation (B3) in Appendix B) and y the mean value of y(k).
Figure 24 illustrates the relation of convergence of the function ‘Fit’ defined in Equation (34) with the
limit value of the identification function for various numbers of order of the ARX model (damage pattern
3; El Centro NS 1940). It can be concluded that the number of order of the ARX model can be determined
on the basis of the convergence property of the function ‘Fit’. The number of order of the ARX model has
been selected as 50–70 in the aforementioned numerical examples. For data with noise (20% of RMS) in
Figure 21, the number of order of the ARX model has been determined appropriately from the function ‘Fit’.

7. CONCLUSIONS
A new method of system identification and damage detection of high-rise buildings has been proposed.
Two major difficulties have been overcome, that is, (i) the physical-parameter system identification in
high-rise buildings and (ii) the identification from limited observation (consecutive floor response just
above and below the target story). The principal results may be summarized as follows:
(1) Stiffness and damping coefficients of a shear-bending model can be determined from the limit
value of an identification function at zero frequency. This identification function can be
constructed from the dynamic equilibrium of the shear-bending model and an approximation
at nonzero frequencies around zero frequency.
(2) Damage detection of a frame model can be conducted in a simple way by using the introduced
identification function and comparing the identification functions for a nondamaged frame and a
damaged frame. Numerical examples demonstrated that, when the damage level of members is fairly
large and the change of story stiffness is relatively significant, the damage detection is possible.
(3) An ARX model is shown to be useful for reliable and accurate identification and damage
detection. Unstable properties of the identification function have been eliminated.
Only a theory of system identification and damage detection has been presented in this paper. The
practical application of the theory to actual situations is necessary. Noise problems in the case of lower
SN ratio may cause some difficulties. This issue will be discussed in the future. It should also be
commented that the present method is applicable only to the data for consecutive two stories. However,
the present method does not require simultaneous measurement of data at all (or many) floors.

Appendix A ARX model [34]

Let k denote the discrete time step. When the output, the input and the white noise are denoted by y(),
u(), and w(), respectively, the ARX model may be described by

Copyright © 2012 John Wiley & Sons, Ltd. Struct. Control Health Monit. (2012)
DOI: 10.1002/stc
M. KUWABARA, S. YOSHITOMI AND I. TAKEWAKI

yðk Þ þ a1 yðk  1Þ þ ⋯ þ ana yðk  na Þ ¼ b1 uðk  1Þ þ ⋯ þ bnb uðk  nb Þ þ wðkÞ (A1)


The transfer function can be expressed in terms of the shift operator q.
GðqÞ ¼ BðqÞ=AðqÞ (A2)
where AðqÞ ¼ 1 þ a1 q1 þ ⋯ þ ana qna (A3)
BðqÞ ¼ b1 q1 þ ⋯ þ bnb qnb (A4)

The relation between the Z transformation and the Fourier transformation is given by Equation (A5)
and the transfer function in terms of the variable q in Equation (A2) can be expressed as the transfer
function in terms of the variable o (circular frequency). In this expression, T0 denotes the sampling period.
q ¼ eioT0 (A5)

b1 eioT0 þ ⋯ þ bn einoT0
G ðo Þ ¼ (A6)
1 þ a1 eioT0 þ ⋯ þ an einoT0

Appendix B Prediction error in ARX model


Define the parameter vector θ and the data vector wðk Þ by Equations (B1) and (B2), respectively. The
prediction of the output at time k from the input–output data until time (k  1) can be expressed by
Equation (B3).
θ ¼ f a1 ⋯ ana b1 ⋯ bnb gT (B1)
 T
w ðk Þ ¼ yðk  1Þ ⋯ yðk  na Þ uðk  1Þ ⋯ uðk  nb Þ g (B2)

^y ðk; θÞ ¼ θT wðk Þ (B3)


With the use of Equation (B3), the prediction error in the ARX model can be described by
eðk; θÞ ¼ yðk Þ  ^y ðk; θÞ ¼ yðk Þ  θT wðkÞ (B4)

Appendix C Least-mean-square estimate of parameter vector


Introduce the objective function by Equation (C1) for predicting the parameter vector θ. Then, the
least-squares method can be applied to the parameter prediction problem as follows.
1 X Nd
JN ðθÞ ¼ e2 ðk; θÞ (C1)
Nd k¼1
In this case, the prediction problem of θ can be reduced to the following simultaneous equations.
Rθ ¼ f; (C2)
where 1  T
R¼ ΦΦT Φ ¼ ½ wð1Þ wð2Þ ⋯ wðNd Þ  (C3)
Nd
1 T   
f¼ Φ y y ¼ yð1Þ yð2Þ ⋯ yðNd Þ gT (C4)
Nd
Therefore, the least-square estimation of the unknown parameters based on the Nd-pair input–output
measured data may be expressed by
^
θ¼ R1 f (C5)

ACKNOWLEDGEMENTS

This research is partly supported by the Grant-in-Aid for Scientific Research (no. 21360267, 23760525) in Japan.
This support is gratefully acknowledged.

Copyright © 2012 John Wiley & Sons, Ltd. Struct. Control Health Monit. (2012)
DOI: 10.1002/stc
SYSTEM IDENTIFICATION AND DAMAGE DETECTION OF HIGH-RISE BUILDINGS

REFERENCES

1. Hart GC, Yao JTP. System identification in structural dynamics. Journal of the Engineering Mechanics Division, ASCE
1977; 103(EM6):1089–1104.
2. Beck JL, Jennings PC. Structural identification using linear models and earthquake records. Earthquake Engineering and
Structural Dynamics 1980; 8:145–160.
3. Hoshiya M, Saito E. Structural identification by extended Kalman filter. Journal of Engineering Mechanics, ASCE 1984;
110(12):1757–1770.
4. Kozin F, Natke HG. System identification techniques. Structural Safety 1986; 3:269–316.
5. Agbabian MS, Masri SF, Miller RK, Caughey TK. System identification approach to detection of structural changes. Journal
of Engineering Mechanics, ASCE 1991; 117(2):370–390.
6. Koh CG, See LM, Balendra T. Estimation of structural parameters in time domain: a substructure approach. Earthquake
Engineering and Structural Dynamics 1991; 20:787–801.
7. Yao JTP, Natke HG. Damage detection and reliability evaluation of existing structures. Structural Safety 1994; 15:3–16.
8. Housner GW, Masri SF, Chassiakos AG(eds).. Proceedings of 1st World Conference on Structural Control. IASC: Los
Angeles, CA, 1994.
9. Hjelmstad KD, Banan Mo R, Banan Ma. Y. On building finite element models of structures from modal response.
Earthquake Engineering and Structural Dynamics 1995; 24:53–67.
10. Ghanem R, Shinozuka M. Structural-system identification I: theory. Journal of Engineering Mechanics, ASCE 1995; 121
(2):255–264.
11. Shinozuka M, Ghanem R. Structural-system identification II: experimental verification. Journal of Engineering Mechanics,
ASCE 1995; 121(2):265–273.
12. Masri SF, Nakamura M, Chassiakos AG, Caughey TK. A neural network approach to the detection of changes in structural
parameters. Journal of Engineering Mechanics, ASCE 1996; 122(4):350–360.
13. Doebling SW, Farrar CR, Prime MB, Shevitz DW. 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.
14. Hjelmstad KD. On the uniqueness of modal parameter estimation. Journal of Sound and Vibration 1996; 192(2):581–598.
15. Housner G et al. Special issue: Structural control: past, present, and future. Journal of Engineering Mechanics, ASCE 1997;
123(9):897–971.
16. Kobori T, Inoue Y, Seto K, Iemura H, Nishitani A (eds). Proceedings of 2nd World Conference on Structural Control. John
Wiley & Sons: Kyoto, 1998.
17. Casciati F (ed). Proceedings of 3 rd World Conference on Structural Control. John Wiley & Sons: Como, 2002.
18. Bernal D, Beck J. Preface to the Special Issue on Phase I of the IASC- ASCE Structural Health Monitoring Benchmark.
Journal of Engineering Mechanics, ASCE 2004; 130(1):1–2.
19. Lus H, Betti R, Yu J, De Angelis M. Investigation of a system identification methodology in the context of the ASCE
benchmark problem. Journal of Engineering Mechanics, ASCE 2004; 130(1):71–84.
20. Johnson E, Smyth A (eds). Proceedings of 4th World Conference on Structural Control and Monitoring, (4WCSCM). IASC:
San Diego, CA, 2006.
21. Nagarajaiah S, Basu B. Output only modal identification and structural damage detection using time frequency & wavelet
techniques. Earthquake Engineering and Engineering Vibration 2009; 8(4):583–605.
22. Fujino Y, Nishitani A, Mita A. Proceedings of 5th World Conference on Structural Control and Monitoring, (5WCSCM).
Tokyo, 2010.
23. Ji X, Fenves GL, Kajiwara K, Nakashima M. Seismic damage detection of a full-scale shaking table test structure. Journal of
Structural Engineering, ASCE 2011; 137(1):14–21.
24. Safak E. Adaptive modeling, identification, and control of dynamic structural systems. I: theory. Journal of Engineering
Mechanics, ASCE 1989; 115(11):2386–2405.
25. Takewaki I, Nakamura M. Stiffness-damping simultaneous identification using limited earthquake records. Earthquake
Engineering and Structural Dynamics 2000; 29(8):1219–1238.
26. Takewaki I, Nakamura M. Stiffness-damping simultaneous identification under limited observation. Journal of Engineering
Mechanics, ASCE 2005; 131(10):1027–1035.
27. Udwadia FE, Sharma DK, Shah PC. Uniqueness of damping and stiffness distributions in the identification of soil and
structural systems. Journal of Applied Mechanics, ASME 1978; 45:181–187.
28. Hernandez-Garcia MR, Masri SF, Ghanem R, Figueiredo E, Farrar CR. An experimental investigation of change detection in
uncertain chain-like systems. Journal of Sound and Vibration 2010; 329(12):2395–2409.
29. Hernandez-Garcia M, Masri SF, Ghanem R, Figueiredo E, Farrar RA. A structural decomposition approach for detecting,
locating, and quantifying nonlinearities in chain-like systems. Structural Control and Health Monitoring DOI: 10.1002/
stc.396.
30. Barroso LR, Rodriguez R. Damage detection utilizing the damage index method to a benchmark structure. Journal of
Engineering Mechanics, ASCE 2004; 130(2):142–151.
31. Takewaki I, Nakamura M. Temporal variation of modal properties of a base-isolated building during an earthquake. Journal
of Zhejiang University SCIENCE A 2009; 11(1):1–8.
32. Maeda T, Yoshitomi S, Takewaki I. Stiffness-damping identification of buildings using limited earthquake records and ARX
model. Journal of Structural Construction Engineering Architectural Inst. of Japan2011; 666:1415–1423(in Japanese)..
33. Mendel JM. Lessons in Estimation Theory for Signal Processing, Communications, and Control, 2nd Edition, Prentice Hall:
Upper Saddle River, New Jersey, 1995.
34. Adachi S. Fundamentals of System Identification. Tokyo Denki University Press: Tokyo, 2009 (in Japanese).

Copyright © 2012 John Wiley & Sons, Ltd. Struct. Control Health Monit. (2012)
DOI: 10.1002/stc

You might also like