Sunteți pe pagina 1din 17

Mechanical Systems and Signal Processing 58-59 (2015) 228244

Contents lists available at ScienceDirect

Mechanical Systems and Signal Processing


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

A condition assessment method for time-variant structures


with incomplete measurements
Y. Ding a,b,n, B.Y. Zhao a,b, B. Wu a,b, G.S. Xu a,b, Q. Lin c, H.B. Jiang a,b, Q.S. Miao d
a
Key Lab of Structures Dynamic Behavior and Control of the Ministry of Education, Harbin Institute of Technology, Harbin 150090,
Peoples Republic of China
b
School of Civil Engineering, Harbin Institute of Technology, Harbin 150090, Peoples Republic of China
c
China Southwest Architectural Design and Research Institute CORP.LTD, Chengdu 610041, Peoples Republic of China
d
Beijing Institute of Architecture Design, Beijing, Peoples Republic of China

a r t i c l e in f o

abstract

Article history:
Received 3 April 2013
Received in revised form
22 November 2014
Accepted 27 December 2014
Available online 17 January 2015

The structural damage incurred in a seismic event is always time-variant. In this paper, a new
time-variant structural system identification method is proposed based on a two-stage strategy
and incomplete structural acceleration responses. In the first stage, an external excitation
identification method is developed for a time-variant structural system. The unknown
structural response could be re-constructed with the average acceleration discrete algorithm
in this stage. In the second stage, structural parameter is identified and updated with a reduced
extended Kalman filter which can improve the computational effort. The re-constructed
structural response and identified external excitation are used in the second stage for the
damage identification and model updating. The proposed method is validated numerically with
the simulation of a fifteen-storey shear frame structure subject to earthquake excitation. A
model of a fourteen-storey concrete shear wall building was also studied experimentally with
shaking table tests to further validate the proposed method. This shear wall building has a twostorey steel frame on top with base isolation. Both the stiffness of the model and the interface
force in the isolator at the bottom of the steel frame during the seismic excitation were
estimated with the proposed method. Results from both numerical simulations and laboratory
tests indicate that the proposed method can be used to identify structural parameters
and external excitations effectively based on a few number of polluted structural acceleration
measurements.
& 2015 Elsevier Ltd. All rights reserved.

Keywords:
Time-variant parameter identification
Force identification
Extended Kalman filter
Regularization

1. Introduction
During the earthquake, a structure may suffer damages and some structural components may behave with time-variant
property. However, most of the cracks caused by earlier severe excitation may get closed, and the effect of the closed cracks
on the structural stiffness may not be fully reflected in the dynamic behavior or the shift of the modal frequencies of
structure when under small environmental excitation. This would adversely influence the post-event structural condition
assessment. Therefore, it is essential to evaluate the condition and the load-bearing capacity of the structural system with
the measured structural response during the earthquake. Accurate time-variant structural parameter identification would

Corresponding author.
E-mail address: dingyong@hit.edu.cn (Y. Ding).

http://dx.doi.org/10.1016/j.ymssp.2014.12.019
0888-3270/& 2015 Elsevier Ltd. All rights reserved.

Y. Ding et al. / Mechanical Systems and Signal Processing 58-59 (2015) 228244

229

be helpful to the post-event evaluation, diagnosis and subsequent maintenance. Commonly, it is difficult to judge when and
where damage occurs based on incomplete response measurements from the structural system and excitation.
Structural parameter identification, model updating and external excitation identification have been actively investigated in
the last two decades. There are mainly three categories of methods for structural parameter identification: frequency domain
methods, time domain methods and methods in the time-frequency domain. The methods in frequency domain include the peak
picking method, methods with transfer function, frequency response function, frequency domain decomposition and
eigensystem realization algorithm, etc. [1]. With the available tools in frequency domain, modal parameters such as natural
frequencies and, mode shapes, mode shape curvature, flexibility matrices, modal strain energy, etc. can be identified. [2].
Methods in time domain include the Ibrahim time domain method, least-squares complex exponential method and the ERA
methods. With measurements of the structure in the time domain, the location and severity of the local structural damage can
be detected [3]. Mixed time and frequency domain techniques, such as the short-time Fourier transform and empirical mode
decomposition, have also been applied to identify the parameters of structural systems [4].
Investigations on the time-variant structural parameter identification have also been conducted on linear and nonlinear
structures over the last two decades [57]. A number of techniques have been proposed for the identification of civil
structures, such as least-squares methods [8,9], Kalman filters [1012], and Monte Carlo methods including particle filters
and other ensemble approaches [1315]. Least-squares methods for damage detection require the velocity or displacement
responses from the structure. These responses are always integrated from the acceleration in practical application, and they
cannot be obtained directly. The implementation of these methods in engineering practices may be difficult with the
contaminated acceleration measurement, and this is a drawback of the least-squares method in identification. The Kalman
filter is an effective means of system parameter identification for a linear or nonlinear structure. An adaptive tracking
technique based on the extended Kalman filter was proposed to identify the structural parameters and their increments
[16]. Methods with extended Kalman filter are suitable to linear structures or structures with small nonlinearity with no
requirements of a priori knowledge of the time of occurrence of the anomalies. Hence, they could be applied to conduct the
condition assessment for the civil infrastructures online. Unscented Kalman filter is another kind of estimation tool
approximated with a Gaussian random variable represented by chosen sampling point but without the calculation of
Jacobians. It performs better than the extended Kalman filter in systems with higher-order nonlinearities [17]. The Monte
Carlo methods including ensemble Kalman filter and particle filters, can solve the inverse problem for the nonlinear systems
with non-Gaussian posterior probability of the state. These Monte Carlo methods can approximate the posterior probability
of the state through the generation of a large number of samples. However, the Monte Carlo methods require a large number
of samples, depending on the number of the factors in the state variable, making the identification process time consuming.
And most existing time-variant parameter identification methods require the information of velocity, displacement or
external force measurement of the structural system. Severe damage or nonlinearity of civil structures after the earthquake
can usually be observed. The vibration-based condition assessment methods for structures mainly detect the structural
damage which cannot be easily noticed with visual inspection. The extended Kalman filter is an alternative and efficient tool
to conduct the rapid structural condition assessment for structures in this case.
The dynamic load evaluation is another crucial tool for structural condition assessment and health monitoring. The
external excitation estimation methods can be classified into categories of direct methods and indirect methods. The force
transducers are installed at locations where the forces are applied with the direct method. In most conditions, indirect
method is used for the evaluation of external forces acting on a structure [18,19] due to the lack of access to the loading
position and the limited number of force transducers. The load environment assessment with indirect method is also an illconditioned inverse problem. A lot of indirect force re-construction or force identification methods have been developed
[20,21]. Many methods have also been proposed for the identification of external excitations, including deterministic forces
[22], stochastic forces [23] and methods based on artificial intelligence [24]. They are usually based on the finite element
model of the structure [25], and some of them could simultaneously identify the structural parameter and excitation [26].
One of the most common assumptions in existing force identification methods is that the structural parameter is timeinvariant. Two-stage identification methods [27,28] have been investigated with the assumption of time-invariant structure
while time-variant structural parameters identification method was investigated with the assumption of known external
excitation [29].
In this paper, a new two-stage system identification method is proposed for the linear time-variant civil structures based
on incomplete acceleration measurements. The work described in this paper enables rapid evaluation of the damaged
structure subjected to the earthquake. In the first stage, an external excitation identification method is proposed for timevariant structural system with average-acceleration discrete algorithm. The unknown structural response is re-constructed
with incomplete measurement. In the second stage, to improve the computational effort a reduced extended Kalman filter is
utilized to estimate the structural parameter with both the re-constructed structural response and the measured structural
response obtained in the first stage. The proposed two-stage identification method was validated numerically with a fifteenstory shear frame structure subject to earthquake excitation. A fourteen-storey concrete shear wall building is investigated
experimentally with shaking table tests to further validate the proposed method. Results from both numerical simulations
and laboratory tests indicate that the proposed method can identify structural parameters and external excitations
effectively based on the structural acceleration responses from only a few sensors. The proposed method is capable to
identify the dynamic load and structural parameters fairly accurately even with measurement noise, practical model error
and environmental disturbances.

230

Y. Ding et al. / Mechanical Systems and Signal Processing 58-59 (2015) 228244

2. Force identication for time-variant structural system


The equation of motion of a damped viscous structural system subject to ground motion can be generally written as
Ctxt
_ Ktxt  MGx g t
M xt

where M, C(t) and K(t) are the mass, damping and stiffness matrices of the structural system respectively and the damping
and stiffness matrices may be time-variant. x g t is the ground acceleration, G is the location matrix of the seismic effect.

_
xt,
xt
and xt are respectively the vectors of acceleration, velocity and displacement responses of the structural system.
Rayleigh damping is assumed for the illustration and damping ratios of the first two modes are taken to be 0.05 and 0.05
respectively for numerical simulation in this paper. It is noted that other damping model could also be adopted for the
proposed method.
2.1. Equation of motion for the time-variant structure in state space
Considering the time-variant stiffness of the structural system, Eq. (1) can be re-written for the continuous state as
M x Ctx_ Ktx LFt

where the stiffness matrix K(t) and the damping matrix C(t) is time-variant, L is the location matrix of the external
excitation and F(t) is the external excitation. The equation of motion of the structural system shown in Eq. (2) can also be
expressed generally in the state space as
z_ t AC tzt BC L  Ft wt
"
#
"
0
xt
where zt
, AC t
_
xt
 M  1 Kt

I
 M  1 Ct


, BC

0
M1


and w(t) denotes the process noise with covariance

matrix Q0 (t). The superscript C denotes matrices for the continuous system. Vector yt is the output of the structural system
and it is assembled from the measurements with Eq. (4).
Rv xt
_ Rd xt vt
yt R a xt
a

where R ,R and R are the output influence matrices with dimensions m  ndof for the measured acceleration, velocity and
displacement respectively, m is the dimension of the measured responses and ndof is the number of DOFs of the structure.
The term v(t) in Eq. (4) denotes the observation noise with the covariance matrix R(t). Eq. (4) can be rewritten as
yt R C zt DC  L  Ft vt
5
h
i
C
1
1
C
1
where R R d R a M Kt Rv  Ra M Ct and D Ra M . In practice, the measurement data is discrete and the
continuous state equation is required to be transformed into discrete equation. Eqs. (3) and (5) can be converted into the
following discrete equations as
D
zj 1 AD
j zj B  L  Fj wj

yj R C zj DC  L  Fj vj j 1; 2; ; N

Superscript D denotes matrices for the discrete structural system. N is the total
of sampling points, dt denotes
 number

the time step. The process noise wj and observation noise vj are setted as wj  N 0; Q 0j and vj  N0; Rj . The output yj is
solved from Eqs. (6) and (7) with zero initial responses in terms of the previous input Fk ; k 0 ; 1 ; ; j and we have
yj

j
X

HD
k  L  Fj  k vj

k0

where HD
k is the Hankel matrix which is different from that in the time-variant discrete algorithms. The following
paragraphs present the force identification method with average acceleration discrete algorithm for the time-variant
structures.

2.2. Average acceleration discrete algorithm


Newmarks method is generally used to evaluate the dynamic response of a structural system. Average acceleration stepby-step integration method is one form of the Newmarks method with good convergence property. The assumptions of the
average acceleration discrete algorithm are


x x k 1
t
x_ k 1 x_ k k
2

Y. Ding et al. / Mechanical Systems and Signal Processing 58-59 (2015) 228244

xk 1 xk x_ k t


x k x k 1
t 2
4

231

where k denote the number of time steps. Rearranging, the incremental acceleration and velocity can be represented
as

x k 1
2x_ k
t 2
x
x_
x k x k 1  x k 4 k 1 4 k  2x k
t
t
x_ k x_ k 1  x_ k 2

10

Substituting Eq. (10) into Eq. (1) at the (k 1)th time instant, the following equation can be obtained as



4
2
2
k Ck M x_ k Kk xk
C
M

M
x
k
k
k
k

1
t
t
t 2

11

It is noted that the mass matrix is assumed constant but the damping and stiffness matrix of the time-variant structure
are changing with time. The incremental displacement can then be represented as



  1

4
2
4
_

C
M
x
x k
M

K
F

M
x


K
x
k
k1
k
k
k
k k
t k
t
t 2

  1

4
2
4

12
M C k Kk
Fk 1 Fk Mx_ k  2Kk xk
t
t
t 2
Substituting Eq. (12) into Eq. (10), the incremental velocity can be represented as:

x_ k x_ k 1  x_ k 2
2

x k
 2x_ k
t 2


  1

1
4
2
4
M C k Kk
Fk 1 Fk Mx_ k  2Kk xk  2x_ k
2
2
t
t
t t

13

The displacement and velocity at the (k 1)th time instant can be represented as the summation of those at the kth time
step and the increments as
"

xk 1
x_ k 1

"

x k x k
x_ k x_ k

#
14

Eq. (14) can be written in state space as


ND
zk 1 AND
k zk Bk L  Fk

15

where
"

#
xk 1
;
zk 1 _
xk 1

2

1

6 t 2 M tCk Kk
7
6 
BND
1 7
k 4
5;
2
4
2
t t 2 M t Ck Kk


1
4
2
Kk
6 I 2 t 2 M t Ck Kk
ND
Ak 6
1
4  4 4
2
Kk
t t 2 M t Ck Kk

4
4
2
t t 2 M t Ck Kk

8
4
M 2t Ck Kk
t 2 t 2

1

1

3
M

M I

7
7:
5

With the assumption of the average acceleration discrete algorithm, output yj can also be written similar to Eq. (8) as
yj

j
X
k0

HND
k  L  Fj  k

16



ND
C
ND
C ND ND
C
 1 ND
BND
with HND
ki
0 RB1 D and Hk R A 1 B1 R
k , where superscript ND denote the system matrix in a new
1 Ai
discrete version with average acceleration discrete algorithm. It is noted from Eq. (16) that the time-variant matrices AND,
BND and H from the average acceleration discrete algorithm in this paper is different from those in existing force
identification methods with the time-invariant structural system. The proposed method allows the force identification for
the time-variant structural system.

232

Y. Ding et al. / Mechanical Systems and Signal Processing 58-59 (2015) 228244

2.3. Force identification of the time-variant structural system


The force identification can be based on Eq. (16) after rewriting it in a general form [28] as
Y m HL F
where

17

H0
6
H
6
1
HL 6
6
4
HN  1

H0

HN  2

7
0 7 S
7L ;
0 7
5
H0

60
6
LS 6
4
0

07
7
7;
05

and Ym denotes the measured responses. Regularization method would provide an improved solution to this kind of
identification equation [30,31]. The damped least-squares method proposed by Tikhonov is adopted to give bounds to the
solution as
HL T Y m HL T HL F

18

where is the non-negative damping coefficient governing the participation of the least-squares error in the solution.
Solving Eq. (18) is equivalent to minimizing the function
JF; HL F  Y m 2 F2

19

The L-curve method proposed by Hansen [30,31] is adopted to find the optimal regularization parameter . A relative
percentage error in the identified forces is defined as
error F

Fid  Ftrue
 100%
Ftrue

20

where Ftrue is the real force acting on the structure and Fid is the identified force. The exact value of external force
Ftrue is difficult to obtain in practice. Therefore, an alternative form of error of identification from measurement can be
represented as
m
re
error acc x  x
m
x  100%

21

where the erroracc denotes the relative error between the measured acceleration and re-constructed acceleration with the
identified forces.
3. Time-variant parameter identication
3.1. Parameter identification with EKF
For the structural parameter identification with reduced EKF, a new discrete state equation is written as Eq. (22) and
observation equation can be written as Eq. (23) with reduced extended Kalman filter.

k k  1 k  1

22

yk hk R a  M  1 Ck x_ k Rk ; xk LFk vk
23

T
where Rk ; xk denotes the resisting force, t 1 ; 2 ; ; n is the unknown parameter vector in general. In the new
state equation, k is the process noise vector with covariance Qk. It is noted that Eq. (22) is only involved in the inverse
analysis with EKF but not the forward dynamic analysis. The parameter k in Eq. (22) denotes the unknown difference

T
of the parameter between each time step. In this paper t 1 ; 2 ; ; n denotes the unknown stiffness change
coefficient or reduction ratio. Considering the reduction of the stiffness according to Eq. (23), the measurement equation can
be written as
yk hk R a  M  1 Ck x_ k K0  Kk  xk LFk vk

24

where yk is the measured structural acceleration. Matrix K0 is the initial stiffness matrix, K is the stiffness change of the
structural system and its representation will be provided in the next Section. The restoring force Rk ; xk can be expressed
as K0  Kk  xk . The parameter identification procedure with EKF can be summarized as follows,
Step 1 Initialize the filter as follows:

^ 0 E0

Y. Ding et al. / Mechanical Systems and Signal Processing 58-59 (2015) 228244

P 0 E0  ^ 0 0  ^ 0 T 

233

25

Step 2 For k1,2,,


(a) Compute the partial derivative matrices for the state equation:

dk  1

Fk 1
I
d
^ k 1

26

(b) Conduct the time update of the state estimation and the error covariance as followswhere I is a unit matrix.
Pk Pk 1 Q k  1


^ k ^ k  1

27

(c) Compute the partial derivative matrix for the measurement equation as

hk

Ra  M  1 Ck x_ k K0  Kk  xk


Hk

^

^ k

k


_
C

x
k
k
x_ k Ck k 
 Ra  M  1
xk  k Kk

28

The matrix Hk will be discussed in Section 3.2.


(d) Perform the measurement update of the state estimate and estimation error covariance as follows
Kk Pk HTk Hk Pk HTk Rk  1



^ ^ K y  h^
k

Pk I Kk Hk Pk

29

3.2. Time response sensitivity matrix


The time response sensitivity with respect to the time-variant structural parameter, matrix Hk in Eq. (28), is calculated in
each discrete time step. If the damage extent at the ith element of the jth time step in the structure is represented as a
i
stiffness change coefficient or reduction factor, j , a change in the global stiffness matrix of the structure can be described as

Ne
X

ij Ki0

30

where Ne denotes the total number of finite elements of the structure, and Ki0 denotes the ith element stiffness in the intact
initial state assembled in the global stiffness matrix.

m 15 c15 k15
m 14 c14 k 14
m 13 c13 k 13
m 12 c12 k12

m3 c 3 k 3
m 2 c2 k 2
m 1 c1 k 1

Fig. 1. Fifteen-storey shear frame.

234

Y. Ding et al. / Mechanical Systems and Signal Processing 58-59 (2015) 228244

With the assumption of Rayleigh damping and performing differentiation to both sides of Eq. (1) with respect to the
i
reduction factor , existing sensitivity method [26] would give
x
x_
x
K
K
M i C i K i  i x a2 i x_

31

x_ and x can be obtained by step-by-step time integration method from Eq. (1). They are then substituted
The responses x,
_ i ,x=i can then be solved using the time integration method. Note
i ,x=
into Eq. (31), and the sensitivity matrices x=
_ i and x=i can be
that the stiffness and damping matrices in Eq. (31) are time-variant, and the sensitivity matrices x=
used in Eq. (28) for the extended Kalman filter Estimation.

4. Implementation procedure
Step 1: Obtain mass, damping, stiffness matrix by structural drawings and structural model updating with environmental
random excitation.
Step 2: Conduct measurement on the structure. The measured data for the simulation studies is obtained from the
solution of Eq. (1).
Step 3: Perform the external excitation identification with Eqs. (17)(19) as shown in Section 2.3.
Step 4: Identify the structural parameters with the proposed EKF algorithm.
Step 5: Update the finite element model of the structure and calculate the difference between the updated structural
response and the measured structural responses. Repeat Step 3 until the following convergence criteria is met
0.5
Real damage ratio

0.4

Identified damage ratio

0.3

Damage ratio

0.2
0.1
0
-0.1
-0.2
-0.3
-0.4
-0.5

10

15

20

25

30

Time (s)

Fig. 2. Abrupt damage identification.

0.5
Real damage ratio

0.4

Identified damage ratio

0.3

Damage ratio

0.2
0.1
0
-0.1
-0.2
-0.3
-0.4
-0.5

10

15

20

25

Time (s)

Fig. 3. Smoothly changed single damage identification on the 1st floor.

30

Y. Ding et al. / Mechanical Systems and Signal Processing 58-59 (2015) 228244


Fi 1  Fi

rTol

Fi 1

235

32

Subscript i denotes the number of global iteration and Tol is a small prescribed values taken as 10  4 for all studies in
this work.
0.5
Real damage ratio

0.4

Identified damage ratio

0.3

Damage ratio

0.2
0.1
0
-0.1
-0.2
-0.3
-0.4
-0.5

10

15

20

25

30

Time (s)

0.5
Real damage ratio

0.4

Identified damage ratio

0.3

Damage ratio

0.2
0.1
0
-0.1
-0.2
-0.3
-0.4
-0.5

10

15

20

25

30

Time (s)

0.5
Real damage ratio

0.4

Identified damage ratio

0.3

Damage ratio

0.2
0.1
0
-0.1
-0.2
-0.3
-0.4
-0.5

10

15

20

25

30

Time (s)

Fig. 4. Smoothly changed multi damage identification: (a) first floor, (b) second floor and (c) third floor.

236

Y. Ding et al. / Mechanical Systems and Signal Processing 58-59 (2015) 228244

5. Numerical simulation studies


Three cases of time-variant parameter identification with a shear structure are numerically investigated. The structure is
assumed linear with the mass of each storey equals 4  105 kg and the storey stiffness of each floor equals 2  108 N/m. The
base excitation is the El-Centro (1940, NS) earthquake ground motion and the excitation is taken as unknown input for the
structure. The sampling rate of measurement is 100 Hz. 30 s of horizontal accelerations at the 1st, 4th, 8th, 12th and 15th
floor floors are taken as the measured responses.
When there is noise in the measured response, the polluted response is simulated by adding a normal random
component to the unpolluted structural responses as

x m x EP N noise x

33

where Ep is the percentage noise level, Nnoise is a standard normal distribution with zero mean and unit standard deviation,
is the standard deviation of the measured response.
x
5.1. Case 1Shear frame with single abrupt damage
The shear frame structure has 15-storeys with rigid base connection as shown in Fig. 1. Local damage is modeled with
10% abrupt reduction of storey stiffness in the 1st floor where the shear force is the largest. The abrupt stiffness reduction is
assumed to occur at 3 s from the beginning of the excitation. Fig. 2 compares the real storey damage ratio and the identified
storey damage ratio time histories of the 1st floors. The time of occurrence and location of the abrupt damage could be
identified fairly accurately with 5% noise in the incomplete structural acceleration measurement. It is shown in Fig. 2 that
the identified damage ratio of the structure in the first floor is close to the real stiffness change though with some
fluctuation. The identified damage ratio in other floors is not listed and the maximum value is less than 3%. The identified
damage extent, the occurrence time and location are acceptable although there is a large error at the beginning of the
stiffness reduction as well as some small fluctuations in the damage ratio time histories.
5.2. Case 2Shear frame with single ramped stiffness reduction
A 10% stiffness reduction on the first floor is assumed to take place over 3 s time duration. The same parameters and setup for
last study are adopted. Fig. 3shows that the identified time history is close to the true one, though with notable small fluctuations
as well as large errors at the beginning of the time history. The initial error at time zero is transient which should diminishes to

Fig. 5. Shear wall building model on shaking table: (a) shear wall building and (b) shear wall building with additional steel frame.

Y. Ding et al. / Mechanical Systems and Signal Processing 58-59 (2015) 228244

237

zero after a few second. The identification results at other time instants are fairly accurate and acceptable. The location, severity
and occurrence time of single time-variant structural damage can be identified with the proposed approach.

5.3. Case 3Shear frame with multi reduction in stiffness


The multiple stiffness reduction is modeled as: 10% abrupt stiffness reduction at the first floor occurs at 3 s from the
beginning of the excitation; 10% ramped stiffness reduction occurs at 5 s within 3 s at the second floor; similar ramped

x
Fig. 6. Floor plan for the concrete structure: (a) floor plan for the lowest five storeys and (b) floor plan for the sixth to fourteenth storeys.

238

Y. Ding et al. / Mechanical Systems and Signal Processing 58-59 (2015) 228244

stiffness reduction at the third floor occurs at 8 s from the beginning. Fig. 4(a)(c) give the identified results with 10%
measurement noise. The feature of the identified damage ratio time histories is similar to those shown in Cases 1 and 2. The
identified damage in 2nd and 3rd floors have time lag while the identified damage in the first floor has small time lag. It is
shown in the Ref. [32] that the damage on different floors did not influence each other with a complete measurement of the
acceleration. In this paper, the incomplete measurement is applied for the identification. So the time lag of the identified
damage on 2nd and 3rd could be explained by acceleration collected at the 1st floor but not on the 2nd and 3rd floor. And
also there are notable small fluctuations in the identified damage ratio time histories. However, the trend of the damage
ratio, time of damage occurrence and the damage severity can be fairly accurately evaluated.
6. Laboratory validation
A scaled 14-storey reinforced concrete shear wall building was studied with earthquake excitation for the validation of
the proposed method. There is an additional two-storey steel frame on top of the concrete building with base isolation. The
scaled model shown in Fig. 5(a) was constructed on the shaking table of the Institute of Engineering Mechanics, China
Earthquake Administration. The acceleration on each floor is recorded as shown in Fig. 5(b) but only a few of them is used
for the condition assessment. The geometric scale ratio of the structure is 1/6 and the floor plans are shown in Fig. 6. Scaled
El-Centro (1940, NS) earthquake excitation is applied as the base excitation in the y-direction (weak direction). The scale
ratios for time and ground acceleration are 0.3 and 1.86, respectively.
The shear wall building model was made from grade M7.5 mortar. The structure was reinforced with grade M15 mortar
and steel wire with an average yield strength of 852.22 MPa. The steel wire for the shear wall has the configuration of
0.8@13  13 in the weak direction (x-direction) and 0.8@13  13 in the strong direction (y-direction). Double layers of steel

Fig. 7. Floor plan for the two-storey steel frame.

Table 1
Comparison of modal frequencies before the excitation.
Order
1
2

Modal frequency
Damping ratio
Modal frequency
Damping ratio

Experimental model

Numerical cantilever model

4.49
1.23%
9.16
1.18%

4.49
1.23%
9.16
1.18%

Y. Ding et al. / Mechanical Systems and Signal Processing 58-59 (2015) 228244
0.15

0.15
Identified damage ratio

Identified damage ratio

0.1

0.1
0.05
Damage ratio

Damage ratio

0.05
0
-0.05
-0.1

0
-0.05
-0.1

-0.15

-0.15

-0.2
-0.25
0

10

-0.2

15

Time (s)

10

15

Time (s)

0.04

0.04
Identified damage ratio

Identified damage ratio

0.02

0.02

0
Damage ratio

Damage ratio

239

-0.02
-0.04
-0.06

-0.02
-0.04
-0.06

-0.08

-0.08

-0.1
-0.12
0

10

-0.1

15

Time (s)

10

15

Time (s)

0.02

0.01
Identified damage ratio

Identified damage ratio

0.01
0

Damage ratio

Damage ratio

0
-0.01
-0.02
-0.03
-0.04

-0.01
-0.02
-0.03

-0.05
-0.04
-0.06
-0.07
0

Time (s)

10

-0.05

15

Time (s)

x 10

15

x 10

Identified damage ratio

Identified damage ratio

-1

Damage ratio

Damage ratio

10

-4

-3

-2
-3

-5

-10

-15

-4
-5
0

Time (s)

10

15

-20

Time (s)

10

15

Fig. 8. Time-variant stiffness identification result of shaking table test: (a) stiffness of 1st floor, (b) stiffness of 2nd floor, (c) stiffness of 3rd floor,
(d) stiffness of 4th floor, (e) stiffness of 5th floor, (f) stiffness of 6th floor, (g) stiffness of 7th floor, (h) stiffness of 8th floor, (i) stiffness of 9th floor, and
(j) stiffness of 10th floor, (k) stiffness of 11th floor, (l) stiffness of 12th floor, (m) stiffness of 13th floor, (n) stiffness of 14th floor.

240

Y. Ding et al. / Mechanical Systems and Signal Processing 58-59 (2015) 228244

-4

x 10

0.005
Identified damage ratio

-0.005

Damage ratio

Damage ratio

Identified damage ratio

-1
-2
-3

-0.01
-0.015
-0.02

-4

-0.025

-5

-0.03

-6

-0.035
0

10

15

Time (s)

10

15

Time (s)
-3

0.005

x 10

Identified damage ratio

Identified damage ratio

-0.005

Damage ratio

Damage ratio

-0.01

-5

-10

-0.015
-15

-0.02
-0.025

-20
0

10

15

Time (s)

15

-3

-3

10

Time (s)

x 10

x 10

Identified damage ratio

Identified damage ratio

Damage ratio

Damage ratio

-2
-4
-6
-8

-2
-4
-6

-10
-8

-12
-14

-10
0

10

15

Time (s)

10

Time (s)

Fig. 8. (continued)

Fig. 9. Cracks at the bottom of the shear wall building.

15

Y. Ding et al. / Mechanical Systems and Signal Processing 58-59 (2015) 228244

241

Damage ratio

0.2
X: 4.375
Y: 0.06291

0.1

X: 5.33
Y: -0.03223

X: 7.315
Y: -0.1178

-0.1
X: 4.84
Y: -0.1239

-0.2
0

10

15

10

15

Storey drift of 2

nd

(mm)

Time (s)
0.6
0.4

X: 5.285
Y: 0.1529

X: 4.235
Y: 0.4762

0.2
0

-0.4

X: 7.155
Y: -0.2676

X: 4.625
Y: -0.3254

-0.2
0

5
Time (s)

Fig. 10. Comparison of the storey drift and damage ratio on 2nd floor.
Table 2
Comparison of modal frequencies after the excitation.
Order Experimental
model

Numerical cantilever
model

1
2

4.28
8.80

4.27
8.72

wire mesh were constructed with 0.8@13  13 and 1.9@25  25 for the floor plate in x-direction and y-direction,
respectively. The two-storey steel frame structure is fabricated from grade Q235 40 mm  60 mm  2 mm rectangular steel
tube, and is fixed to the top of the 14-storey shear wall building with bolts through the base plates of columns. The
photograph of the steel frame can be seen in Fig. 5(b) and the floor plan of the steel frame is shown in Fig. 7. Each storey of
the steel frame is 483.3 mm high. Each column of the frame is welded at the bottom to a base plate which has four 20 mm
bolt holes for connecting to the concrete roof of the building as shown in Fig. 7. The storey stiffness of the model and the
interface force in the isolation at the bottom of the steel frame during the seismic excitation are estimated with the
proposed method.
The whole model weighs 7.62 t and 678 kg additional mass is added on each floor level to simulate the inertia effect from
floor mass. The mass of the structure is mainly found at each floor level, and the 14-storey shear wall building and twostorey steel frame could be simplified into a lumped mass cantilever structure connected with 16 beam elements. The modal
property of the simplified model and the shear wall building structure before the excitation is compared in Table 1. The
modal frequencies and modal damping are found very close together. The model error between the practical structure and
numerical structure in lower modes are small enough to conduct the structural condition assessment. Accelerometer model
941B made by the Institute of Engineering Mechanics, China Earthquake Administration and Data Acquisition System model
6000DAS are used in the shaking table test. The sampling rate is 200 Hz and 13 s horizontal accelerations at the 6th, 10th,
and 13th floor are collected for the storey stiffness identification.
In the experimental study, the earthquake excitation is taken as known input while the interaction force between the
shear wall structure and the additional steel frame at the top is taken as unknown input. The proposed structural system
identification method is applied. The identified time-variant damage ratios at the first floor to the 14th floor are shown in
Fig. 8. It is noted from the time history of the damage ratio that during the earthquake excitation, the damage of structure is
noted accumulating with time. The largest damage is at the first floor as shown in Fig. 8(a), which is consistent to the visual
inspection of the crack pattern in the model as shown in Fig. 9. There are some fluctuations in the time history of the storey
damage ratio as shown in Fig. 8, which may be due to structural damage from the opening and closing of cracks in the
concrete structure. As shown in Fig. 8(a)(f), the structural damage is mainly in the lowest six storeys, which is also
consistent with the visual inspection. The comparison of the stiffness reduction and the storey shift of the second floor is
shown in Fig. 10. The cracks of the concrete structure during the earthquake excitation are open-closed. It is shown in Fig. 10
that the local maximum and minimum damage ratio occurred right after the peaks of the storey drift with a time delay.
The first two modal frequencies of the numerical model and experimental model after the earthquake are compared in
Table 2. The error in the identification result is very small. The comparison of the calculated and measured accelerations is

242

Y. Ding et al. / Mechanical Systems and Signal Processing 58-59 (2015) 228244

0.2
Measured response
Calculated response

Acceleration (m/s2 )

0.15
0.1
0.05
0
-0.05
-0.1
-0.15
-0.2
-0.25

10

15

Time (s)

0.25
Measured response
Calculated response

0.2

Acceleration (m/s2 )

0.15
0.1
0.05
0
-0.05
-0.1
-0.15
-0.2
-0.25

10

15

Time (s)

0.3
Measured response
Calculated response

Acceleration (m/s2 )

0.2
0.1
0
-0.1
-0.2

10

15

Time (s)
Fig. 11. Comparison of the simulated and measured acceleration . (a) Acceleration on 6th floor, (b) Acceleration on 10th floor & (c) Acceleration on 13th floor.

shown in Fig. 11. Though there are some errors between the measured and simulated responses from the damaged
structure, yet the time histories are very close together.
The horizontal interaction between the shear wall building and the two-storey steel frame are identified with Steps 1 to
3 in Section 4. The horizontal shear force could also be evaluated experimentally as
Fis msf x sf csf x_ sf ksf xsf

34

where sf denotes the DOFs of the steel frame and Fis is the shear force in the base isolation. In the interface force calculation
as in Eq. (34), the two storey steel structure is simplified as two-storey lumped mass structure. The stiffness of the steel
structure is calculated with the geometry and material information of the design drawing. The damping matrix csf is also
calculated with Rayleigh damping assumption. The displacement and acceleration on the steel structure is measured on the
storey level. The velocity is integrated from the measured acceleration. The calculated shear force from Eq. (34) is compared
with the identified force in Fig. 12. The identified force has the same pattern as the calculated shear force in the isolation
indicating that the identified result is fairly acceptable for practical application.

Y. Ding et al. / Mechanical Systems and Signal Processing 58-59 (2015) 228244

243

Shear force from the isolation (N)

3000
Calculated shear force
Identified shear force

2000
1000
0
-1000
-2000
-3000

10

12

Time (s)

Fig. 12. Comparison of the identified external force and calculated external force.

7. Conclusions
A new method based on the extended Kalman filter estimation and average acceleration discrete algorithm is proposed
for the inverse identification of structural parameter and external excitation of a time-variant structure. Numerical studies
with a 15-storey shear frame and experimental validation with a scaled model of a reinforced concrete building are
conducted. The external force and local damage in the structure can be identified fairly accurately with polluted
measurements and model error. The recursive computational technique adopted in this study can be used for the dynamic
response calculation with the proposed method. With experiences gained from this study, it may be concluded that the
proposed two-stage structural condition assessment method could be used for the external force identification and model
updating simultaneously in future research for nonlinear structural condition assessment. Though the results of the
proposed method are acceptable, this kind of methods for other damping assumption including modal damping need to be
developed in the future study. Also, the innovative identification methods including particle filters and ensemble techniques
instead of reduced EKF would be further investigated.

Acknowledgements
The work described in this paper was supported by Project no. 01319406 Supported by Natural Scientific Research
Innovation Foundation in Harbin Institute of Technology, Projects nos. 51161120360, 51308160 and 91315301 of National
Natural Science Foundation of China and Projects nos. 2013M541383 and 2014T70342 of China Postdoctoral Science
Foundation.
References
[1] J.N. Juang., Applied System Identification, Prentice Hall, Englewood Cliffs, 1994.
[2] R. Brincker, L. Zhang, P. Andersen., Modal identification of output-only systems using frequency domain decomposition, Smart Mater. Struct. 10 (3)
(2001) 441445.
[3] X.Q. Zhu, S.S. Law, Damage detection in simply supported concrete bridge structures under moving vehicular loads, J. Vib. Acoust. 129 (1) (2007)
5865.
[4] S. Nagarajaiah, B. Basu., Output only modal identification and structural damage detection using time frequency & wavelet techniques, Earthquake
Eng. Eng. Vibr. 8 (2009) 583605.
[5] G. Kerschen, K. Worden, A.F. Vakakis, J.C. Golinval., Past, present and future of nonlinear system identification in structural dynamics, Mech. Syst. Sig.
Process. 20 (3) (2006) 505592.
[6] G. Kerschen, M. Peeters, J.C. Golinval, A.F. Vakakis., Nonlinear normal modes, Part I: A useful framework for the structural dynamicist, Mech. Syst. Sig.
Process. 23 (1) (2009) 170194.
[7] M. Peeters, R. Viguie, G. Srandoura, G. Kerschen, J.C. Golinval., Nonlinear normal modes, Part II: Toward a practical computation using numerical
continuation techniques, Mech. Syst. Sig. Process. 23 (1) (2009) 195216.
[8] A.W. Smyth, S.F. Masri, A.G. Chassiakos, T.K. Caughey, On-line parametric identification of Mdof nonlinear hysteretic systems, J. Eng. Mech. 125 (2)
(1999) 133142.
[9] J.N. Yang, S. Pan, S. Lin., Least square estimation with unknown excitations for damage identification of structures, J. Eng. Mech. 133 (1) (2007) 1221.
[10] Masaru Hoshiya, Etsuro Saito, Structural identification by extended Kalman filter, J. Eng. Mech. 110 (12) (1997) 17571770.
[11] A. Corigliano, S. Mariani., Parameter identification in explicit structural dynamics: performance of the extended Kalman filter, Comput. Meth. Appl.
Mech. Eng. 193 (3638) (2004) 38073835.
[12] S.J. Julier, J.K. Uhlmann, H.F. Durrant-Whyte., A new method for the nonlinear transformation of means and covariances in filters and estimators, IEEE
Trans. Autom. Control 45 (3) (1995) 477482.
[13] E.N. Chatzi, A.W. Smyth., The unscented Kalman filter and particle filter methods for nonlinear structural system identification with non-collocated
heterogeneous sensing, Struct. Control Health Monit. 16 (2009) 99123.
[14] J. Ching, J.L. Beck, K.A. Porter, R. Shaikhutdinov., Bayesian state estimation method for nonlinear systems and its application to recorded seismic
response, J. Eng. Mech. 132 (4) (2006) 396410.
[15] L. Li, H. Ji, J. Luo. (2006). The iterated extended Kalman particle filter, in: Proceedings of the International Symposium on Communications and
Information Technologies, Beijing, China, 2005, pp. 11721175.

244

Y. Ding et al. / Mechanical Systems and Signal Processing 58-59 (2015) 228244

[16] S. Haykin, A.H. Sayed, J.R. Zeidler, P. Yee, P.C. Wei., Adaptive tracking of linear time-variant systems by extended RLS algorithms, IEEE Trans. Signal
Process. 45 (5) (1997) 11181128.
[17] M. Wu, A. Smith, Real-time parameter estimation for degrading and pinching hysteretic models, Int. J. Nonlinear Mech. 43 (9) (2008) 822833.
[18] H. Inoue, Review of inverse analysis for indirect measurement of impact force, Appl. Mech. Rev. 54 (6) (2001) 503524.
[19] T.H.T. Chan, C. OConnor, Wheel loads from highway bridge strains, J. Struct. Eng. ASCE, 116, , 1990, 17511771.
[20] M.F. Green, D. Cebon., Dynamic response of highway bridges to heavy vehicle loads: theory and experimental validation, J. Sound Vib. 170 (1) (1994)
5178.
[21] T. Kucharski., A method for dynamic response analysis of time-variant discrete systems, Comput. Struct. 76 (4) (2000) 545550.
[22] B.J. Dobson, E. Rider., A review of the indirect calculation of excitation forces from measured structural response data, Proc. Inst. Mech. Eng. Part C J.
Mech. Eng. Sci. 204 (1990) 6975.
[23] D.M. Trujillo, H.R. Busby, Practical Inverse Analysis Engineering, CRC Press, New York, NY, 1997.
[24] T. Uhl., The inverse identification problem and its technical application, Arch. Appl. Mech. 77 (5) (2007) 325337.
[25] M.F. Green, D. Cebon., Dynamic response of highway bridges to heavy vehicle loads: theory and experimental validation, J. Sound Vib. 170 (1) (1994)
5178.
[26] Z.R. Lu, S.S. Law, Identification of system parameters and input force from output only, Mech. Syst. Sig. Process. 21 (5) (2007) 20992111.
[27] Y. Lei, Y.Q. Jiang, Y. Liu. (2012). A two-step Kalman estimation approach for the identification of nonlinear structural parameter, In: The 12th
International Symposium on Structural Engineering, November 17-19, 2012, Wuhan, China.
[28] S.S. Law, Y. Ding., Substructure methods for structural condition assessment, J. Sound Vib. 330 (15) (2011) 36063619.
[29] S.S. Law, Y. Ding., Structural damping identification based on an iterative regularization method, J. Sound Vib. 330 (10) (2011) 22812298.
[30] P.C. Hansen, Analysis of discrete ill-posed problems by means of the L-curve, SIAM Rev. 34 (4) (1992) 561580.
[31] A.M. Tikhonov., On the solution of ill-posed problems and the method of regularization, Sov. Math. 4 (1963) 10351038.
[32] J.N. Yang, S. Lin., An adaptive extended Kalman filter for structural damage identification, Struct. Control Health Monit. 13 (2003) 849867.

S-ar putea să vă placă și