Anda di halaman 1dari 8

Journal of Information & Computational Science 12:11 (2015) 4457–4464 July 20, 2015

Available at http://www.joics.com

A Modified Optimal Sensor Placement Method Based on


the Effective Independence Method ?

Xiaohe Xu, Youfa Yang ∗, Shuai Li, Dian Xu


School of Civil Engineering, Chongqing University, Chongqing 400045, China

Abstract

In this paper the effective independence method based on the Fisher information matrix is modified.
Taking capacity factor into account, deformation energy of specific stiffness was introduced to modify
assignment matrix. Theoretical formula of the effective independence method, kinetic energy method
and three estimating criteria were deduced. Then, with the background of a steel truss, the above
optimal approaches were used for sensor placement and their effects were evaluated according to the
three estimating criteria. At last, the results showed that the modified effective independence method
was better than the others methods above. It could reflect structural character better.

Keywords: Optimal Sensor Placement; The Effective Independence Method; Kinetic Energy Method;
Modified Effective Independence Method; Steel Truss; Estimating Criteria

1 Introduction
Non-destructive testing technology based on vibration is a major branch of structural damage
detection. NDT technology is used to detect structural damage existing, damage location and
damage severity by structural dynamic characteristics which coming through dynamic testing.
But, we can only arrange sensors on structural key position to get reliable and total modal
information because of limitation of test condition [1]. So, it is important to arrange sensor
placement in structural dynamic testing.
Optimal sensor placement methods have been studied early overseas. Kammer presents the EFI
(effective independence) method in [2] in study of big space structure’s sensor placement. Guyan’s
model polycondensation is another optimal sensor placement method, mainly by removing the
degrees of freedom which produces less modal response and retaining the degrees of freedom
which produces more modal response as the layout of the sensor [3]. Heo presents a new optimal
method based on kinetic energy, which puts structural kinetic energy of measure point as optimal
objective [4]. CUI-FEI presented modal confidence to arrange sensors optimal in [5].
?
Project supported by the Fundamental Research Funds for the Central Universities (No. 106112014CD-
JZR200006).

Corresponding author.
Email address: yfyang6296@163.com (Youfa Yang).

1548–7741 / Copyright © 2015 Binary Information Press


DOI: 10.12733/jics20106119
4458 X. Xu et al. / Journal of Information & Computational Science 12:11 (2015) 4457–4464

In this paper, a modified effective independence method is presented. The EFI method, KEM
with the modified EFI method in a steel truss case are compared. Then, three comparison
criteria including Modal Assurance Criteria (MAC), least mean square error criterion and the
Fisher information matrix criterion are used to compare the above three optimal sensor methods.
It results that linear independence of the EFI method is greater than that of KEM. The optimal
result of EFI method is better than that of KEM. The improved method taking energy into
account can overcome the shortcomings which may miss some of the less energy measurement
point of the EFI method, and the improved method can reflect more original structure character
because of the more linear independence.

2 Theory of Optimal Sensor Placement


2.1 The Effective Independence Method
The core ideology of the EFI method ranks the candidate sensor locations according to their
contribution to the linear independence of the target modal partitions. In an iterative manner,
the locations which do not contribute significantly are removed.
It is assumed that the output response at the sensor locations is Us , so the output response
equation can been written as Eq. (1).
N
X
Us = Φs q = φi qi (1)
i=1

in which q is the target modal response, Φs is a n × N matrix of finite element model target mode
shapes partitioned to the sensor degrees of freedom, n is number of freedom, N is the modal
order, φi is the ith mode shape, and qi is participation factor of mode shape. The least squares
solution of q can been written as follows
£ ¤−1 T
q̂ = ΦTs Φs Φs Us (2)

Taking noise S into account, the output response Us is reflected by Eq. (3)
N
X
Us = Φs q + S = φi qi + S (3)
i=1

An efficient unbiased estimator yields an estimate error covariance matrix P of the form
£ ¤
P = E (q − q̂)(q − q̂)T = Q−1 (4)
1 1
Q = 2 ΦTs Φs = 2 A0 (5)
σ σ

where Q represents the Fisher information matrix. Maximization of Q results in the maximization
of the A0 , so we can build matrix E in the form
£ T ¤−1 T
E = Φs A−1 T
0 Φs = Φ s Φ s Φs Φs (6)
X. Xu et al. / Journal of Information & Computational Science 12:11 (2015) 4457–4464 4459

2.2 The Kinetic Energy Method


Different from the EFI method, KEM is based on modal kinetic energy of all candidate sensor
locations, and the kinetic energy can been calculated with the expression

KE = ΦT M Φ (7)

The result of CHOLESKY decomposition for mass matrix is

M = V TV (8)

We subject Eq. (8) into Eq. (7), so

KE = ΦT V T V Φ = (V Φ)T V Φ (9)

It is assumed that Ψ = V Φ, so KE is given by

KE = ΨT Ψ (10)

Then using an iterative manner as the EFI, the locations which do not contribute significantly
model kinetic energy are removed. The termination condition of the iteration is number of sensor
equal to number of allowing sensors.

2.3 The Modified Effective Independence Method


The traditional optimal sensor placement methods may lead to structural information losses
without regarding modal kinetic energy. Now a new modified way taking modal kinetic energy into
account is much more efficient. This algorithm using driving-point residue coefficient reflecting
the weights of the kinetic energy updates the effective independence method. So the EFI method
assignment matrix E can been expressed as
m
£ T ¤−1 T X φ2ij
EC = E · Ci = Φs Φs Φs Φs · (11)
ω
j=1 ij

where ωij is the jth mode frequency of ith element, and Φij is the jth mode shape of ith element.

3 Estimate Criterion of Optimal Sensor Placement


In order to compare the modified EFI with the others two optimal sensor methods, three different
estimate criterions are proposed. The first criterion is the Modal Assurance Criterion (MAC),
the second criterion is based on the mean square error calculated with both the FE model and
the cubic spline interpolated mode shapes. The third criterion is based on the Fisher information
matrix.
4460 X. Xu et al. / Journal of Information & Computational Science 12:11 (2015) 4457–4464

3.1 The Modal Assurance Criterion


The traditional modal assurance criterion is a fine tool for estimating intersection angle of mode
shapes, and it can reflect relativity of two space vector, which can been calculated by the equation
as follows
¡ T ¢2
φ φj
M ACij = T i ¡ T ¢ (12)
(φi φi ) φj φj
in which, φi and φj are the ith and jth mode shape. The value of MAC is in the range of
0 ≤ M AC ≤ 1. If value of MAC is 0, it means that the two vectors are orthogonal. If value of
MAC is 1, it means that the two vectors are completely related. So the bigger MAC value is, the
vectors have more interrelated, and it reflects less vibratory character of structure. The smaller
MAC value is, the vectors have less interrelated, and it can reflect much more vibratory character
of structure.

3.2 The Least Mean Square Error Criterion


Taking a steel truss for example, we can get structure mode shape through ANSYS as a group of
theoretical data. Another group of empirical data can been get by way of cubic spline interpolated
mode shapes. Then mean square error can be expressed as follows
1 P n ¡ ¢2
N Φm − Φsij
X σi j=1 ij
σ= (13)
i=1
n
in which, Φm s
ij is ith mode shape got by cubic spline interpolated, Φij is ith mode shape got by
numerical analysis, and σi is standard deviation of each order mode shape. The smaller the mean
square error is, the better effective the sensor optimal placement is.

3.3 The Fisher Information Matrix Criterion


This criterion is used to estimate two groups of mode shapes interrelated. We use the ratio of
determinant of the Fisher information matrix after deleting candidate sensor and the original
Fisher information matrix. It can been written as
det (Q0 )
η= × 100% (14)
det (Q0 )

where Q0 is the original Fisher information matrix, Q0 is the Fisher information matrix after
deleting candidate sensor, so the bigger is the value of η, the more effective sensor placement is.

4 Numerical Example and Estimate of Optimal Sensor


Placement
In this paper, with the background of a steel truss, the above three optimal approaches are used
for sensor placement. The Finite Element Modeling of steel truss is built by ANSYS software. In
X. Xu et al. / Journal of Information & Computational Science 12:11 (2015) 4457–4464 4461

this case, link 1 element is chosen for plane truss element, each member bar has the same section,
with 0.01 m2 area, and its density is 7800 kg/m3 , Poisson’s ratio is 0.3, young’s modulus E is 210
GPa. Plane size of truss can be seen in Fig. 1. Each node of truss has two direction’s degree of
freedom. For example, the DOF of node 1 be written as 1x and 1y, and so on.

7 8 9 10 11

2000
1 2 3 4 5 6
2000 2000 2000 2000 2000
10000

Fig. 1: Drawing of finite element modal of truss

In this case, the first five mode shapes and frequencies of structure are drawn. Due to space
limitations drawing of the first fourth mode can be seen in Fig. 2.

8 7 9 8 10
7 6 9 11

12 13 14 15 16 17
11 18
10 19
Y 3 3 4
2 2 4 5
1 5 6
Z X
(a) The first mode (f1=41.509 Hz)

7 9 8 10 9
7 6 8 11

13 14 15 16 17 18
10 11 12 19
Y
3 3 4 4 5
1 2 2 5 6
Z X
(b) The second mode (f2=75.391 Hz)

6 8
7 7
9
8 11
12 10 9
11 13
10 14
15 19
Y 2 16 17 18
2 3
1
Z X 3 6
4 4 5
5
(c) The third mode (f3=129.98 Hz)
9
7 8
8 10 9
7 6 11
14 15
13 16
10 11 12 17 18 19
Y
3 3 4
Z X 1 2 4 6
2 5
5
(d) The fourth mode (f4=220.75 Hz)

Fig. 2: The first fourth modal shapes and frequencies of truss


4462 X. Xu et al. / Journal of Information & Computational Science 12:11 (2015) 4457–4464

4.1 The Optimal Results of Three Sensor Placement Methods


The above three approaches are used to arrange sensor locations, and all the best results for the
12, 10 and 8 sensor locations are listed and compared in Table 1-3.

Table 1: Comparison of the optimal sensor locations in 12 measured point’s case


Optimal results of DOF

KEM 2y 5y 7x 8x 9y 6x 8y 10y 3y 4y 11x 5x


EFI 5y 7x 8x 9y 6x 2y 8y 4y 5x 10y 3y 4x
Modified EFI 5y 7x 8x 2y 9y 6x 8y 10y 4y 3y 5x 11x

Table 2: Comparison of the optimal sensor locations in 10 measured point’s case


Optimal results of DOF

KEM 6x 9y 2y 5y 7x 8x 11x 10y 8y 3y


EFI 5y 9y 6x 7x 8x 8y 2y 5x 4y 10y
Modified EFI 5y 9y 6x 7x 2y 8x 8y 5x 4y 10y

Table 3: Comparison of the optimal sensor locations in 8 measured point’s case


Optimal results of DOF

KEM 9y 5y 6x 2y 7x 8x 11x 8y
EFI 6x 5y 9y 7x 8y 8x 4y 2y
Modified EFI 6x 5y 9y 7x 2y 8x 8y 4y

4.2 Estimate of Optimal Results


The Modal Assurance Criterion (MAC), the least mean square error criterion and the Fisher
information matrix criterion are used to compare optimal results in arranging 8 sensor placement
case.

1) The Modal Assurance Criterion

Values of the three above methods’ MAC are calculated by Eq. (12). The results are listed in
Table 4 and Fig. 3 (a). As shown in the results, we know that the modified effective independence
method has the least MAC value and it can better reflect structural vibratory character.

2) The Least Mean Square Error Criterion

A group of theoretical data of FEM is calculated by ANSYS; another group of empirical data is
got by way of cubic spline interpolated after drawing eight points’ mode shapes. The results of
X. Xu et al. / Journal of Information & Computational Science 12:11 (2015) 4457–4464 4463

Table 4: The value of MAC’s comparison of above three approaches


Optimal Value of MAC Mean square
approaches error of MAC
1, 2 mode 2, 3 mode 3, 4 mode 4, 5 mode

KEM 0.002 0.0029 0.0072 4.88e-2 0.0247


EFI 0.0091 3.86e-5 0.0115 0.0035 0.0075
Modified EFI 2.902e-6 0.001 3.70e-3 0.0086 0.0047

Table 5: The least mean square error’s comparison of above three approaches
Mean square error of each mode Total mean
Optimal
square error
approaches The first The second The third The fourth The fifth
mode mode mode mode mode

KEM 0.0238 0.0452 0.0093 0.0277 0.0162 0.0273


EFI 0.0124 0.0085 0.0037 0.0062 0.0055 0.0079
Modified EFI 0.0082 0.0014 0.0057 0.0106 0.0021 0.0066

mean square error for these two groups data are calculated by equation (13) and listed in Table
5 and Fig. 3 (b).
As shown in Table 5 and Fig. 3 (b), the value of mean square error of modified EFI is 0.0066,
which is the least. It shows that optimal results of modified EFI is better than the other ap-
proaches.
Mean square error value

0.06 0.06
Value of MAC

0.04 0.04

0.02 0.02

0 0
Mean square 5
EFI 4,5 error EFI 4
KEM KEM b r
e
3,4 3
e r num
Modified EFI 2,3 es Modified EFI 2 ord
Mo d dal
1,2 1 Mo
(a) (b)

Fig. 3: The MAC and least mean square error’s comparison of above three approaches

3) The Fisher Information Matrix Criterion

From Fig. 4 we can see that with the increasing number of sensor locations, the ratio of the Fisher
information matrix is growing for the above three approaches. Results cure of EFI method is close
to that of modified EFI method. The two optimal methods are much more effective in sensors
placement than KEM.
4464 X. Xu et al. / Journal of Information & Computational Science 12:11 (2015) 4457–4464

Ratio of fisher information matrix determinant


1.0
KEM
EFI
0.8 Modified EFI

0.6

0.4

0.2

0
6 8 10 12 14 16 18 20 22
Number of sensors

Fig. 4: The ratio of the Fisher information matrix

5 Conclusions
The effective independence method ranks the candidate sensor locations according to their contri-
bution to the linear independence of the target modal partitions. The candidate locations which
do not contribute significantly are removed. KEM method stresses in kinetic energy of candidate
sensor, which builds information matrix taking kinetic energy into account. The modified EFI
method considering kinetic energy is based on EFI, so the new method overcome the shortcoming
of EFI method and can much effectively reflect the structure character.

References
[1] Shanshan Wang, Qingwen Ren, A study on optimal sensor placement for structural modal param-
eters testing [J], Journal of Dynamics and Control, 3(1), 2005, 67-71
[2] D. C. Kammer, Sensor placement for on orbit modal identification of large space structures [J],
Journal of Guidance, Control, and Dynamics, 14(2), 1991, 252-259
[3] R. J. Guyan, Reduction of stiffness and mass matrices [J], AIAA Journal, 3(2), 1965, 380-392
[4] G. Heo, M. L. Wang, D. Satpathi, Optimal transducer placement for health monitoring of long
span bridge [J], Soil Dynamics and Earthquake Engineering, 16, 1997, 495-502
[5] Yaxun Yang, Xianwu Hao, Lei Sun, Optimal placement of sensors for a bridge structure based
on energy coefficient-effective independence method [J], Journal of Vibration and Shock, 29(11),
2010, 119-123
[6] Thomas G. Carne, Clark R. Dohman, A modal test design strategy for modal correlation [C],
In: Proc. 13th International Modal Analysis Conference, New York: Union College, Schenectady,
1995, 927-933
[7] Z. K. Peng, P. W. Tse, F. L. Chu, An improved Hilbert-Huang transform and its application in
vibration signal analysis [J], Journal of Sound and Vibration, 286(1/2), 2005, 187-205
[8] Xiaodan Sun, Jinping Ou, Optimal sensor placement based on sensitivity of damage parameters
[J], Journal of Harbin Institute of Technology, 42(10), 2010, 1530-1535

Anda mungkin juga menyukai