You are on page 1of 17

Vibration Control of Composite Plates via Optimal

Placement of Piezoelectric Patches


S. T. QUEK,* S. Y. WANG AND K. K. ANG
Department of Civil Engineering, National University of Singapore, 1 Engineering Drive 2 #07-03, Singapore 117576
ABSTRACT: A simple optimal placement strategy of piezoelectric sensor/actuator (S/A)
pairs for vibration control of laminated composite plate is presented, where the active
damping effect under a classical control framework is maximized using the finite element
approach. The classical direct pattern search method is employed to obtain the local optimum,
where two optimization performance indices based on modal and system controllability are
studied. The start point for the pattern search is selected based on the maxima of integrated
normal strains consistent with the size of the collocated piezoelectric patches used. This would
maximize the virtual work done by the equivalent actuation forces along the strain field of an
initial state. Numerical simulation using a cantilevered and a clamped composite square plate
illustrate the effectiveness of the proposed strategy, where the results coincide with the global
optimal layout from exhaustive search for both modal and system controllability indices. The
number of trials to reach the optimal position is few compared to an initial blind discrete
pattern search approach. The reduced control effectiveness of both plates under constrained
input voltages is illustrated.
Key Words: placement optimization, vibration control, composite plates, piezoelectric sensors
and actuators, pattern search method, global optimum

INTRODUCTION
HE specific problem of optimal placement of piezoelectric sensors/actuators in a smart structure has
aroused wide interest among the researchers from many
different disciplines because it can be crucial to the
robust functioning of active control systems (Padula and
Kincaid, 1999). A large number of publications from
widely different engineering disciplines on optimal
sensor/actuator placement methods have been reviewed
by several researchers (Padula and Kincaid, 1999;
Frecker, 2002). Generally, the optimization problem
can be classified as continuous or discrete and the
optimization methods would be either non-systematic
(such as some intuitive cut and try placement
techniques) or systematic optimization methods such
as Simulated Annealing (SA), Tabu Search (TS), and
Genetic Algorithms (GAs).
As an intuitive non-systematic optimization method,
Hwang et al. (1994) suggested an intuitive design
strategy for a cantilevered laminated composite plate
with piezoelectric sensor/actuator (S/A) pairs by using
the finite element (FE) method. It was observed that the
S/A pairs placed close to the fixed end are most effective
for all modes whereas those positioned at the free end
are least effective. S/A pairs placed away from, but

*Author to whom correspondence should be addressed.

JOURNAL

OF INTELLIGENT

MATERIAL SYSTEMS

symmetric with respect to, the center lines are preferable


in torsional vibration control. However, these observations are based on computation results at a few selected
discrete locations only and systematic optimization
methods were not introduced to obtain the optima for
these S/A pairs.
Kim and Ko (1998) studied the optimal design of
smart structures on which piezoelectric sensors and
actuators are integrated with simple control logic to
reduce the sound radiated from the structure at
resonance and off-resonance frequencies using the FE
method. For the constrained optimization problem, the
sequentially unconstrained minimization technique was
used and Powells method was applied to find the
minimum point. However, as observed by some
researchers (Hamda et al., 2002), the remeshing technique can be a source of numerical noise. Powells
method, like all the other classical local search (or hill
climber) methods, can only give fairly good result but
can easily be trapped at a local optimum which may be
far way from the global one. It is well known that the
results from the classical local search methods would
depend heavily on the start point or initial guess.
The most popular systematic optimization methods
nowadays are SA, TS and GA. Systematic methods can
lead to significant better results than the nonsystematic
methods for the placement optimization problem if
carefully implemented. Kincaid and Padula (2002) used
the TS approach to seek k most effective locations to
AND

STRUCTURES, Vol. 14April/May 2003

1045-389X/03/4/5 022917 $10.00/0


DOI: 10.1177/104538903034686
2003 Sage Publications
Downloaded from jim.sagepub.com by kishore bondada on December 29, 2014

229

230

S. T. QUEK

control and/or sense vibration of a given truss structure.


It was reported that their search approach dominates
the traditional approaches of finding optimal designs.
TS is one of the efficient methods for large problems and
it converges rapidly with quality solutions. However, the
Tabu length has to be tuned properly for each target
problem. The SA method, which starts with a
Metropolis Monte Carlo simulation at a high temperature, is also adopted to solve the optimal placement
problem (Onada and Hanawa, 1992; Liu et al., 1997).
However, the balance between the maximum step size
and the number of Monte Carlo steps depends very
much on the characteristics of the search space or energy
landscape, and thus the convergence speed may become
too slow (Trosset, 2001). GA, a stochastic search
method that mimics the metaphor of natural biological
evolution, has also been proven to be effective in solving
this placement optimization problem. Gao et al. (2000)
employed GA to solve the problem of optimal discrete
placement of piezoelectric patches and presented a new
selection factor based on immune diversity. The
improved GA approach is shown to be efficient.
However, the computational cost of GA is usually
high for large problems and the GA operator must be
handled carefully to avoid premature convergence.
Generally, global optimization remains an important,
practical yet unanswered problem in optimization
theory in terms of implementation (Papalambros and
Wilde, 2000). The traditional enumeration method is
applicable only if the problem size is small. Among most
optimization methods, perhaps the most effective means
is based on good initialization or start point, followed
by gradient optimization. Ideally, it would be desirable
to select the start point in the neighborhood of the
global optimum and take the local optimum found as
the global optimum. This approach is possible only if
the nature of the problem is well understood a priori.
Therefore, it is applicable only to specific problems. The
aim of this paper is to apply this approach to finding the
optimal placement of collocated piezoelectric S/A pairs
in order to most effectively suppress the free vibration of
a composite laminated plate using a velocity feedback
control algorithm.
In the present study, the placement of collocated
piezoelectric S/A pairs in a laminated composite plate is
investigated using the FE method resulting in a discrete
optimization formulation. Two performance functions
based on the modal controllability and system controllability are proposed as indices for damping out the free
vibration. A good initial guess to increase the chance of
obtaining a global optimum solution efficiently is
essential. This paper suggests using the sum of normal
strains of the composite plate under an initial strain
field, where the virtual work done by the input voltage
can be approximately maximized, as the starting layout
in conjunction with the direct pattern search method.

ET AL.

Local optimal solutions of this discrete placement


optimization problem can thus be found by using a
direct pattern search. A cantilevered and a clamped
square composite plates are used to illustrate numerically the efficiency of the proposed approach and the
nature of the solutions with respect to modal and system
controllability. The constraints on actuator voltages in
practice will be considered and the effect studied.

FORMULATIONS FOR VIBRATION CONTROL


USING PIEZOELECTRIC PATCHES
For simplicity, a regular FE mesh is used. In this light,
a discrete integer coordinate naming system consistent
with the element boundaries is adopted. The coordinates
of a collocated piezoelectric S/A pair is chosen to denote
the left bottom corner point of the pair, as shown in
Figure 1. The S/A pairs are assumed to occupy fully the
area of an element or group of elements and bonded on
the top surface of the host composite plate. These S/A
pairs do not overlap each other. Therefore, the
placement of the S/A pairs is limited to discrete
locations and the design variables are the discrete
integer coordinates of the S/A pairs. The equations of
motion in the absence of structural damping with
respect to the reduced modal space can be written as
(Sun et al., 1998; Wang et al., 2001)
   
f g fg F B fa g

where fg 2 Rr
1 is the modal coordinate vector; r the
number of retained
modes;  2 Rr
r the matrix
 of
 
r
1

the reduced force vector; B 2
eigenvalues; F 2 R
Rr
p the actuator influence matrix; fa g 2 Rp
1 the
actuator voltage vector; and p the total number of
piezoelectric S/A pairs.
y
8
7
Pair i
6

xi
5
4
Pair 1

x1

yi

y1
2

1
1

Figure 1. Coordinates defining location of S/A pair based on FE


mesh.

Downloaded from jim.sagepub.com by kishore bondada on December 29, 2014

231

Vibration Control of Composite Plates

In the present study, the classical constant gain


negative velocity feedback control method is adopted
and a constant gain matrix G is used to couple the
input actuator voltage vector fa g and the output sensor
voltage fs g as (Wang et al., 2001)
fa g Gfs g

 
fs g Gc Kds  X_

where

 
in which Gc is the constant gain matrix of the charge
 
amplifier; Kds the mechano-electronic matrix; and X_
the structural velocity vector. Since the piezoelectric
sensors are configured as strain rate sensors, the
relationship between the output charge Qi and voltage
si of the ith sensor is
si Gic

dQi
dt

where Gic is the constant gain of the charge amplifier of


the ith electroplated sensor. From Gauss law, the charge
output Qi t can be expressed in terms of spatial
integration of the electric displacement Dz over its
surfaces as
i

Q t

Qj t

Z
A
j

Dz dA

Dz dA

A
j

in which the subscript j denotes the element number of



the sensor; A
j , Aj the top and bottom surfaces of the
sensor element; and Dz the electric displacement along
the thickness direction. The latter is given by
Dz e31 "1 e32 "2 e34 "4 e35 "5 e36 "6

where e3i is the dielectric constant along the thickness


direction and "i the corresponding mechanical strain.
In this study, the sensor is made of PVDF or PZT
materials where the dielectric constants are such
that e34 e35 e36 0. Hence, Equation (7) can be
simplified as
Dz e31 "1 e32 "2

In many applications, the response of a structure is


strongly dominated by a specific mode of vibration.
Hence, one of the simplest possible optimal performance
indicators is based on modal controllability (Ryall,
2001), where the piezoelectric S/A pairs are placed to
control a particular mode most effectively. It is assumed
that maximizing the active damping ratio of this
dominant mode of the transient vibration is effective
in maximizing the active damping effect of the global
system. The performance index can be written as
J1 i

where the active damping ratio of the ith mode i is used


to measure the effectiveness in controlling this mode.
A second optimal performance indicator considered
in this study is based on the influence matrix of
actuators, from the viewpoint of controllability (Sun
et al., 1998). This is in view of the fact that regardless of
the control algorithms adopted, a necessary condition
for successful vibration control is that the control forces
exerted by the piezoelectric actuators should affect
all the vibration modes of interest (Ip and Tse, 2001).
To allow for the maximum controllability of only a few
modes up to the total number of the actuators, the
indicator can be written as
p
r
X

2 X
J2 R B p
b2ij
Ri

where
1
Qj t
2

INDICATORS FOR OPTIMALITY

ip1

10

j1

 
where B p is the p
p matrix comprising
  the entries in
the first p lines of the influence matrix B , corresponding to the first p modes; R the weight for the controllability of the first p modes; Ri the weight for the
controllability of each of the other retained modes; and
of the retained modes in
bij the corresponding element

the influence matrix B .
In practice, the voltages of the S/A pairs cannot
exceed their breakdown voltages. This constraint on the
input actuator voltages can be combined with Equation
(10). The optimization problem can be expressed
in simple terms as minimising f z, where f z
f z1 , z2 , . . . , zp J2 and zj refers to the position of the
jth actuator defined by the x and y coordinates shown in
Figure 1. The vector z of the design variables is subject
to the inequality constraints as
g i z < 0

i 1, 2, . . . , p

11

where

Downloaded from jim.sagepub.com by kishore bondada on December 29, 2014

gi z i z i

12

232

S. T. QUEK

in which i z is the input voltage of the ith actuator and


i the allowable voltage for this actuator.
According to Suh and Radcliffe (1978), this
constrained minimization problem can be converted
into an artificial unconstrained problem. Assuming an
interior inequality penalty, the combined modified
objective function (or performance index) U z can be
constructed as

U z f z 

p
X
1
g z
i1 i

13

where 1=gi z is the inverse interior gi z 6 0 inequality


function and  the penalty parameter (0< 1). In
general, the interior penalty method starts at feasible but
suboptimal points and iterates to optimality as  ! 0.
Other forms of the modified objective function with
exterior inequality penalty or equality penalty can be
found in Suh and Radcliffe (1978).
It should be noted that these optimization performance indices cannot be expressed as functions of
the design variables (i.e. coordinates of the S/A pairs
as defined in Figure 1) in an explicit way. Hence,
appropriate optimization methods should be adopted
where for most methods, the selection of the initial guess
(or start point) is extremely important to ensure
convergence to the global optimum. In general, the
global optimum is deduced by obtaining all the local
optima but this would be impractical due to the high
computational cost. Ideally, it would be desirable to
select the start point in the neighborhood of the global
optimum and take the local optimum found as the
global optimum. In the present study, it is proposed
that the good start point be based on an initial strain
field of the dynamic system.

START POINT FOR OPTIMISATION BASED


ON NORMAL STRAINS
If the effect of the mass and stiffness of the S/A pairs
on the dynamic behavior of the composite substrate is
negligible, according to the pin-force model (Barboni
et al., 2000), the equivalent actuator forces are the
constant linear distributed moments along the edges for
different placement. Thus the values of the
 non-zero
elements of the actuator influence matrix B remains
constant for different placement, though the positions
will change with the actual placement. According to
Equation (1), the equivalent actuator forces are maximized by maximizing the input voltages of the actuators
fa g. Based on the constant gain negative velocity
feedback control adopted in this study where the output
of the piezoelectric sensors is coupled to the input of the

ET AL.

piezoelectric actuators according to Equation (2), the


corresponding sensor voltage fs g should also be
maximized, i.e.
fa gmax ! fs gmax

14

To maximize the sensor voltage fs g for the actively


damped free vibration, it can be inferred from Equations (4) and (5) that these strain-rate piezoelectric
sensors should be placed in regions where more electric
charges and high electric currents can be obtained, i.e.

si


max


! Qi max

15

To maximize the induced electric charge, the integration


of electric displacement Dz according to Equation (6)
should be maximized. For PVDF or PZT sensor,
the electric displacement Dz of Equation (8) can be
written as
Dz e31 "1 e32 "2


m2 e31 n2 e32 "x m2 e32 n2 e31 "y

16

in which m cos , n sin  and  is the skew angle of


the sensor, which is constant for a given sensor rather
than a design variable in this study. Hence,


Qi m2 e31 n2 e32

Z



"x d m2 e32 n2 e31

Z


"y d
17

where  refers to the top and bottom surfaces of the


sensor. However, only the bonding surface is considered
in the present study to find the optimal placement. This
approximation would not change the evaluation of the
optimal placement since the thickness of the PVDF or
PZT sensor is small compared with that of the host
composite substrate. Although Equation (17) indicates
that  affects the electric charge, it is prescribed and not
a design variable for the present study.
Therefore, to obtain the maximum input voltages,
Equations (14), (15) and (17) indicate that the surface
bonded sensors should be placed in regions where high
integrated values of the normal strains over the bonding
surface are located. However, the integrated value is a
function of time for a dynamic system. In the present
study, the initial state when the maximum displacements
as well as the maximum strains occur for the case of free
vibration is adopted.
For vibration control of smart piezoelectric composite
plates, isotropic biaxial PVDF/PZT is often used
(e31 e32 ) and thus the induced electric charge in

Downloaded from jim.sagepub.com by kishore bondada on December 29, 2014

233

Vibration Control of Composite Plates

Equation (17) would not be changed by the skew angle


and can be simplified as


max

!


"x "y d

18

Substituting Equations (19) and (16) into Equation (20)


yields


Z
Z


"x d m2 e32 n2 e31
"y d
W~ ai m2 e31 n2 e32


max

21
This implies that to maximize the input voltage of
actuators fa g, the isotropic PVDF/PZT pairs should be
placed in the regions where the sum of the normal
strains integrated over the actuator patch is high.
To replicate the induced mechanical strains generated
by a piezoelectric actuator under an external electric
voltage, the simplified pin-force model is often used
(Crawley and de Luis, 1987; Main et al., 1994; Barboni
et al., 2000). The actuators are assumed perfectly
bonded to the structure, with negligible mass compared
to the mass of the host structure. Usually, the thickness
of the piezoelectric S/A pairs is small (less than 1/10)
compared with that of the host laminated composite
substrate and does not add significantly to the bending
stiffness. Hence, the effect of the piezoelectric actuators
can be approximated quite accurately as equivalent to a
surface shear force (Crawley and de Luis, 1987). This
model is adopted in the present study to derive a good
initial guess for the optimal placement of piezoelectric
S/A pairs.
For a specified applied voltage and assuming
constant electric field along the thickness direction of
a surface-bonded PVDF/PZT actuator, distributed
shear forces are induced at the bonding surface. Based
on linear variation in potential and first-order plate
theory, the equivalent surface shear forces for the
ith PVDF/PZT actuator can be written as (Wang et al.,
2001)
T 1 e31 E3 ti e31 ai
T 2 e32 E3 ti e32 ai

19

where T 1 and T 2 are the equivalent surface shear


forces under the applied voltage ai ; E3 the electric
field along the thickness direction; and ti the thickness
of the ith actuator. Both T 1 and T 2 are proportional
to ai . They serve to actively damp out the vibration
energy of the dynamic system. To maximize the
active damping effect, the actuator should be located
where the virtual work W~ done by T 1 and T 2 along
the surface strain field is maximum at the aforementioned initial state. This virtual work can be
written as
W~

Z


T 1 "1 T 2 "2 d

20

Comparing Equation (21) with Equation (17) indicates that to maximize the virtual work for a given S/A
pair, the actuator should be placed at the location where
the electric charge of the sensor is also the maximum.
Because the PVDF/PZT S/A pair is assumed collocated,
this requirement is guaranteed. Therefore, the collocated
S/A pairs should be located in regions where the normal
strains integrated over the bonding surfaces is high,
resulting in maximum virtual work done by the
actuators. If point sensors/actuators are used, this
approximation is consistent with the statement made
by Crawley and de Luis (1987) that the piezoelectric
actuators should be placed in regions of high average
strains and away from areas of zero strains. This
provides a reasonably good initial guess in obtaining
the global optimal placement of collocated S/A pairs.
The next step involves adopting an appropriate optimization method based on the discrete formulation of the
problem.

LOCAL OPTIMUM BASED ON DIRECT


PATTERN SEARCH
In FE formulation, explicit evaluation of partial
derivatives of the objective function in terms of the
design variables given by the location coordinates of the
S/A pair is almost impossible. This puts most of
the classical methods in a disadvantageous position
(Barboni et al., 2000). A suitable method for the present
discrete optimization FE formulation is the direct
pattern search method (Cooper and Steinberg, 1970).
Partial derivatives of the objective function are evaluated solely based on the values of the objective function
plus information gained from earlier iterations.
The pattern search comprises two types of basic move
(Cooper and Steinberg, 1970). The start point, which is
the first base point, can be derived from the proposed
initial guess based on the surface strain field of an initial
state. At the base point, an exploratory move is made. It
is a restricted univariate search such that a move of very
small length is taken in a single coordinate direction.
Exploratory moves are made with respect to each
component of z. The information gained from the
exploratory moves is then combined to provide the
direction for the pattern move such that there is a high
probability of achieving a higher value of the objective
function. By evaluating points along the selected

Downloaded from jim.sagepub.com by kishore bondada on December 29, 2014

234

S. T. QUEK

direction, the best value is found and the location


adopted as the next base point. The process is repeated
until a local optimum is reached.

NUMERICAL EXAMPLES
Based on the above, a FE code using the C
language and MATLAB m-files was developed.
Numerical examples are presented to illustrate the
effectiveness of the proposed initial guess for the direct
pattern search with either one or two S/A pairs. The
objective is to maximize the modal controllability or
system controllability of the first two modes. A
cantilevered square plate is first presented followed by
the same plate clamped on all edges.
Discrete Optimization of Cantilevered
Smart Composite Plate
The piezoelectric composite plate (Wang et al., 2001),
comprising four T300/976 graphite-epoxy composite
substrate and two outer PZT G1195N layers shown in
Figure 2 is considered. The stacking sequence of the
substrate is antisymmetric angle-ply. The thickness of
each substrate layer is 0.25 mm and the thickness of each
PZT layer is 0.1 mm.
The composite substrate is discretised using a 8
8
mesh with each collocated piezoelectric S/A pair of size
(50
50 mm) spanning 2
2 elements, as shown by the
shaded area in Figure 3. The control gain is assumed to
be 20,000. It should be noted that for the discrete
optimization problem, the optimization results are
actually mesh dependent and more accurate results can
be found using a finer mesh with the optimal placement
found using a coarse mesh as the new start point.
Optimal placement of the S/A pairs to maximize the
modal controllability or system controllability of the
first two modes is done by using the proposed strategy
to obtain the start point and up to two S/A pairs are
considered.

ET AL.

Optimal Placement Based on Controllability


of First Mode
By using the FE code developed (Wang et al., 2001),
the first mode shape and the corresponding contour of
the sum of normal strains are obtained as shown in
Figures 4 and 5, where the modal matrix is normalized
as kxi k 1 for the ith eigenvector xi . It can be seen that
the first mode is a bending mode and the maximum
value of the sum of normal strains is located in a narrow
zone near the fixed end. Slight deviations from
symmetry is expected since the substrate is an angleply antisymmetric laminated composite.
Optimal placement of a single S/A pair is first
investigated. From the sum of normal strains at each
Gauss point of each element of the composite substrate,
the strains are integrated over four elements, consistent
with the size of a single S/A pair. Figure 6 gives the
location with the maximum integrated value and
constitutes the start point for the pattern search. The

Fixed end
i

xi

yi
x

1
1

Figure 3. Model and layout of cantilevered plate with S/A pair.

Fixed end

0.15

20.6mm

Sensor
Actuator

20.1mm

Deflection

200mm
0.1
0.05
0

45

A
2100mm

-0.05
45

40.25mm
45

8
6

45

Cross Section

Figure 2. Geometry of cantilevered composite plate.

4
Pair coordinate y

6
4
2

Figure 4. Shape of first mode.

Downloaded from jim.sagepub.com by kishore bondada on December 29, 2014

Pair coordinate x

235

Vibration Control of Composite Plates

discrete optimal placement based on the pattern search


yields the same location (7, 4) as the local optimum. This
suggests that both the initial guess and the speed of
convergence of the optimal search are excellent. To
damp the first mode most effectively, the S/A pair must
be located near the center line along the fixed end.
The relationship between the local optimum shown in
Figure 6 and the global optimum is also investigated.
Figure 7 displays the relationship between the damping
ratio of the first mode and the discrete location of a S/A
pair. It can be seen that the placement of a pair can
affect the damping ratio significantly. A pair close to the
fixed end is most effective in damping out the first mode
whereas it is least effective at the free end. This
conclusion agrees well with the observation made by
Hwang et al. (1994), in which a formal optimization
approach was not employed in the solution process.
Fixed end

Figure 7 indicates that the global maximum damping


ratio will be reached when the S/A pair is placed at (7, 4)
for the 8
8 mesh, which is similar to the local optimum
obtained in Figure 6, showing the effectiveness of the
proposed start point.
Optimal placement of two S/A pairs is next investigated. The proposed start location is shown in Figure 8
and direct pattern search yields the optimal position in
Figure 9. The optimal location is near the center line of
the fixed end and in the neighborhood of the location of
the initial guess. The difference in the start and optimal
position may be due to the effect of the mass and
stiffness of S/A pairs on the dynamic response, which
may be more significant for two S/A pairs than the case
of one S/A pair.
The relative convergence speed (in terms of the
number of moves required) of the pattern search
method by using the proposed start point is also
investigated, as shown in Table 1. For simplicity, the
ratio between the total number of possibilities for the
discrete placement of the S/A pairs and the total number

0
-5E-005
-0.0001
-0.00015

0.012

-0.0002

Centre line

-0.00025

0.01

-0.00035
-0.0004
-0.00045
-0.0005
-0.00055

Damping ratio

-0.0003

0.008

Fixed end

0.006
0.004
0.002
0
8
8

Figure 5. Contour of sum of normal strains.

4
Pair coordinate y

4
2

Pair coordinate x

Figure 7. Damping ratio for various location of a S/A pair (first


mode).

Fixed end
7

Fixed end

5
1
5

2
3

1
1

Figure 6. Start point and optimal location (coincident) of a S/A pair


(first mode).

Figure 8. Start position for 2 S/A pairs.

Downloaded from jim.sagepub.com by kishore bondada on December 29, 2014

236

S. T. QUEK

ET AL.

y
Fixed end
0.01

Deflection

0.005

Fixed end

-0.005

-0.01
9

7
5

1
1

Pair coordinate y

9
7
Pair coordinate x

3
1

Figure 10. Shape of second mode.

Figure 9. Optimal position for 2 S/A pairs.


Fixed end

Table 1. Relative moves required by discrete pattern


search (first mode).
Number of
S/A Pairs
1
2

Maximum
Damping
Ratio (%)

Total
Number of
Possibilities n1

Total
Number
of Moves n2

Ratio
n1 =n2

1.20
1.89

49
930

4
16

12.25
58.13

0.001
0.0008
0.0006
0.0004
0.0002

Centre line

0
-0.0002

of actual moves made, including exploratory and


pattern moves in the pattern search is used as a relative
measure. It can be seen that few moves are required than
by using the enumeration method (total number of
possibilities) as indicated by high ratio values. It can
also be noted that higher damping effect can be achieved
by using more S/A pairs.
Optimal Placement Based on Controllability
of Second Mode
The shape of the second mode and the corresponding
contour of the sum of normal strains at each node are
shown in Figures 10 and 11. It can be seen that the
second mode is a torsional mode and there are two
concentrated zones for the sum of normal strains, which
are near the fixed end but away from and almost
symmetric with respect to the center line.
Optimal placement of one S/A pair is first investigated. Figure 12 displays the start location and the
discrete optimal position is found to be at the same
location (7, 7) supporting the proposed strategy of using
the maximum strain location for the start point in
providing an efficient solution. The optimal placement
of one S/A pair to damp out the second mode most
effectively is near the fixed end, but away from the
center line, where the maximum integrated value of the
sum of the normal strains is located.
Figure 13 displays the variation of the damping ratio
of the second mode for various discrete locations of a

-0.0004
-0.0006
-0.0008
-0.001
-0.0012

Figure 11. Contour of sum of normal strains (second mode).

Fixed end

3
x

1
1

Figure 12. Start position and optimal location (coincident) of a S/A


pair (second mode).

Downloaded from jim.sagepub.com by kishore bondada on December 29, 2014

237

Vibration Control of Composite Plates

S/A pair, indicating that the S/A position can affect the
damping ratio of the second mode significantly. A single
S/A pair near the fixed end and away from the center
line of the cantilevered plate can be most effective in
damping out the torsional mode. This conclusion also
agrees well with the observation made by Hwang et al.
(1994). Figure 13 indicates that the optimal placement
obtained earlier yields the global maximum damping
ratio and illustrates the effectiveness of the proposed
method in obtaining the solution for active vibration
control against the torsional mode.
Optimal placement of two S/A pairs for damping
out the second mode is next investigated. Figure 14
shows the optimal placement at (7, 1) and (7, 7) which
coincides with the start position, confirming the
efficiency of the proposed approach. The optimal
placement of two S/A pairs to damp out the torsional
mode most effectively is near the fixed end and away

-3

x 10
8

Damping ratio

6
4
2
0
-2
8

from the center line, where two concentration zones of


high values of the sum of normal strains lie.
The relative convergence speed of the pattern search
method by using the proposed start point and direct
pattern search is shown in Table 2, which is indeed
excellent. The results also indicate that the maximum
damping ratio increases with the number of optimally
placed S/A pairs.
Optimal Placement Based on System Controllability
Optimal placement based on the system controllability comprising the first two modes is investigated using
the performance indicator in Equation (10). The
cantilevered plate was set into vibration by a suddenly
removed concentrated force of 1 N at the tip (1, 5) in the
8
8 mesh. The corresponding deformed shape of the
plate at the start time t 0 is shown in Figure 15.
Comparing with Figure 4, it can be seen that the first
mode is the dominant mode of the dynamic response.
The contour of the sum of normal strains at the start
time is shown in Figure 16. The sum is higher in a
narrow region near the fixed end similar to that for the
first mode in Figure 5. Based on the contour, the
predicted optimal placement for two S/A pairs is shown
in Figure 17.
In this simulation, it is assumed that r 30, p 2,
and Ri 1 3 i 30. The value of the weight R is
chosen as R 1
104 to emphasize the importance of

6
4
Pair coordinate y

6
4
2

Table 2. Relative moves required by discrete pattern


search (second mode).

Pair coordinate x

Figure 13. Damping ratio for various position of a S/A pair (second
mode).
Number of
S/A Pairs

Maximum
Damping
Ratio (%)

Total
Number of
Possibilities n1

Total
Number of
Moves n2

Ratio
n1 =n2

0.76
1.63

49
930

3
5

16.33
186

1
2

-3

x 10

Fixed end
Deflection (m)

Fixed end

4
2
0
-2

1
1

6
8

4
Pair coordinate y

6
2

Figure 14. Start position and optimal location (coincident) of 2 S/A


pairs (second mode).

Pair coordinate x

Figure 15. Deformed shape of plate at t 0.

Downloaded from jim.sagepub.com by kishore bondada on December 29, 2014

S. T. QUEK

the first two modes. Discrete pattern search is performed


based on the proposed start point shown in Figure 17
and the corresponding optimal placement is found to be
at the same location. Comparing Figure 17 with 9,
optimal placement based on the controllability of the
first two modes is slightly different from the optimal
placement based on the controllability of the first mode
only, due to the contribution of the second mode. The
corresponding non-dimensional tip deflection w t
wt=w0 at point (1, 5) in the 8
8 mesh is shown in
Figure 18 and the time history of the input voltages of
the S/A pairs in Figure 19. The free vibration is damped
out within 1.5 s. The S/A pair 2 has a higher peak input
voltage of 24.99 V since this pair is placed closer to the
center line of the plate where higher sum of normal
strains is obtained.

ET AL.

If the allowable input voltages for the S/A pairs


are prescribed as 1 2 17 V, discrete pattern search
can be performed using the combined modified objective
function U z in Equation (13), which is found to
1
0.8
0.6
Nondimensional tip deflection

238

0.4
0.2
0
-0.2
-0.4
-0.6
-0.8

Fixed end
-1

0.5

1.5

Time (second)

Figure 18. Non-dimensional tip deflection of cantilevered plate


(unconstrained case).
4E-005
2E-005
0
-2E-005

25

(a)

-4E-005

20 ( )
1 max=20.70

-6E-005
-8E-005

Centre line

Volt

15

-0.0001
-0.00012

Actuator voltage(Volt)

-0.00014
-0.00016
-0.00018
-0.0002

10
5
0
-5
-10

Figure 16. Contour of sum of normal strains at t 0.

-15
-20

0.5

1.5

Time(second)
25

(b)

(2)max=24.99

Volt

20

15
Actuator voltage (Volt)

Fixed end
2

10
5
0
-5
-10

-15
-20

0.5

1.5

Time(second)

Figure 17. Start position and optimal placement (coincident) of 2 S/A


pairs (two modes).

Figure 19. Required input voltages of two S/A pairs (unconstrained


case): (a) Input voltage of S/A pair 1; (b) Input voltage of S/A pair 2.

Downloaded from jim.sagepub.com by kishore bondada on December 29, 2014

239

Vibration Control of Composite Plates

(a)

20

(1)max=16.58

15

Volt

2
7

Fixed end

Actuator voltage (Volt)

10

-5

-10

1
-15

1
1

0.5

1.5

Time(second)

(b)

Figure 20. Optimal placement for 2 S/A pairs with voltage


constraints.

20
15 (2)max=16.00

Volt

Actuator voltage (Volt)

10

1
0.8

Nondimensional tip deflection

0.6
0.4

-5

0.2
-10

0
-15

-0.2

0.5

1.5

Time(second)

-0.4
-0.6
-0.8
-1

Figure 22. Required input voltages of two S/A pairs (constrained


case): (a) Input voltage of S/A pair 1; (b) Input voltage of S/A pair 2.
0

0.5

1.5

Time(second)

Figure 21. Non-dimensional tip deflection of cantilevered plate


(constrained case).

Discrete Optimization of Clamped


Smart Composite Plate

converge to the constrained minimum when


0 <  1
10 12 . As shown in Figure 20, the optimal
location is near the fixed end but away from the center
line, where the sum of normal strains are smaller so that
the required input voltages of the S/A pairs can satisfy
the constraints of input voltages.
The corresponding non-dimensional tip deflection at
(1, 5) is shown in Figure 21 and the time histories of the
input voltages of the two S/A pairs in Figure 22.
Comparing Figure 21 with 18 and 22 with 19, the active
damping is less effective with voltage constraints. The
maximum input voltages of the former S/A pairs are
1 max 16:58 V and 2 max 16:00 V, and are not
exactly 17 V due to the discrete formulation of the
location. These values are significantly lower than the
unconstrained case where the maximum input voltages
are 1 max 20:70 V and 2 max 24:99 V.

The same plate in Diserete Optimization of


Cantilevered Smart Composite Plate is considered for
the case where the four edges are clamped, as shown in
Figure 23, but with control gain taken as 10,000.
The first mode shape of this square plate without any
S/A pair and the corresponding contour of the sum of
normal strains are shown in Figures 24 and 25. The sum
of normal strains is symmetric about the diagonals of
the plate. The maximum positive values occur at the
center whereas the maximum negative values are at the
center of the four edges, as shown in Figure 25.
Figure 26 displays the optimal location of a single S/A
pair to damp out the first mode most effectively. The
local optimum coincides with the start point indicating
the suitability of the sum of normal strains in obtaining
a good start position for the pattern search. The optimal
location is at the center rather than at the edges because

Downloaded from jim.sagepub.com by kishore bondada on December 29, 2014

240

S. T. QUEK

ET AL.

Fixed end

Fixed end

i
5

xi

yi
x

1
1

1
1

Figure 23. FE model of clamped square plate with S/A pair.

Figure 26. Start point and optimal location (coincidient) of a S/A pair
(first mode).

-3

x 10

0.05

0.04
Damping ratio

Deflection

6
4
2

0.03
0.02
0.01

0
0

-2
7

8
8

6
Pair coordinate y 4
2

Patch coordinate y

Pair coordinate x

Patch coordinate x

Figure 27. Damping ratio for various location of a S/A pair (first
mode).

Figure 24. Shape of first mode.

Fixed end

0.0025
0.002
0.0015
0.001
0.0005
0
-0.0005
-0.001
-0.0015
-0.002
-0.0025
-0.003
-0.0035

Figure 25. Contour of the sum of normal strains (first mode).

the positive sum of strains occupies a much larger area


at the center of the plate (see Figure 25). The integrated
value for the given S/A pair is larger, which means that
larger equivalent actuator force can be mobilized to
control the dynamic response.
Figure 27 shows the damping ratio of the first mode
for various locations of one S/A pair. It can be seen that
there are some local optima near the fixed edges
although the global optimum is at the center of this
plate, confirming the results of Figure 26.
For the case of two S/A pairs for damping the first
mode, the local optimum coincides with the start point
confirming the conclusions from previous cases presented. The optimal location is shown in Figure 28.
The second mode shape of the clamped square plate
and the corresponding contour of the sum of normal
strains are shown in Figures 29 and 30. It can be seen
that this mode is a bending mode with two half waves
along the direction of y axis, with the two absolute

Downloaded from jim.sagepub.com by kishore bondada on December 29, 2014

241

Vibration Control of Composite Plates


Fixed end

0.0045
0.004
0.0035
0.003
0.0025
0.002
0.0015
0.001
0.0005
0
-0.0005
-0.001
-0.0015
-0.002
-0.0025
-0.003
-0.0035
-0.004
-0.0045
-0.005

Fixed end
1

2
x

1
1

Figure 28. Start point and optimal placement (coincident) of two S/A
pairs (first mode).

Figure 30. Contour of the sum of normal strains (second mode).

(a)
-3

x 10
5

Deflection

7
0

Fixed end

5
1
3

-5
8
8

6
4
Pair coordinate y

4
2

Pair coordinate x

Figure 29. Shape of second mode.

maxima of the sum of normal strains near the first and


third quarter position.
Figure 31 displays the optimal location of one S/A
pair to achieve the maximum damping ratio of the
second mode. The proposed start point is very close to
but not coincident with the local optimum. This
reinforces the suitability of the proposed method to
obtain a good start point to obtain the optimum
solution efficiently.
The damping ratio of the second mode versus the location of the piezoelectric S/A pair is shown in Figure 32.
It can be observed that the global optima may lie in the
two regions symmetric with respect to the diagonals of
the plate due to the symmetry of the plate and the mode
shape. For this case, the solution is non-unique if one
S/A pair is used.
If two S/A pairs are used to control the second mode,
the optimal location is shown in Figure 33, where the
optimal location coincides with the proposed start point.

(b)

Fixed end

3
x

1
1

Figure 31. Optimal placement of a S/A pair (second mode): (a) Start
point (b) Local optimum.

It can be noted that Figure 33 is not the same as


Figure 32 (which is only for one pair) because the
mass and stiffness of the S/A pairs are considered.
The results indicate the advantage of using two S/A

Downloaded from jim.sagepub.com by kishore bondada on December 29, 2014

242

S. T. QUEK

ET AL.
Fixed end

0.08

Damping ratio

0.06
0.0045
0.004

0.04

0.0035
0.003

0.02

0.0025
0.002

0.0015

-0.02
8

0.001
0.0005

4
Pair coordinate y

-0.000 5

4
2

-0.001

Pair coordinate x

-0.001 5

Figure 32. Damping ratio for various location of a S/A pair (second
mode).

Figure 35. Contour of the sum of normal strains under central point
load.

Fixed end

5
1
3
x

1
1

Figure 33. Start point and optimal location (coincident) for two S/A
pairs (second mode).

-3

x 10
5
4
Deflection (m)

-0.002

3
2
1
0
-1
8
8

6
Pair coordinate y

4
2

4
2

Pair coordinate x

Figure 34. Deformed shape of clamped plate under a central point


load.

pairs simultaneously, which are to be placed around


the maxima of the half waves to maximize the
controllability of the second mode, consistent with
Figure 29.

To investigate system controllability, the clamped


plate was set into vibration by a suddenly removed
concentrated force of 100 N at the central point (5, 5).
The corresponding deformed shape of the clamped plate
at the start time is shown in Figure 34. The dominance
of the first mode is exhibited by the similarity of the
deformed shape under free vibration with the first mode
shape (Figure 24). The contour of the sum of normal
strains at the start time is shown in Figure 35, similar to
that for the first mode in Figure 25. It is again symmetric
about the diagonals due to the symmetry of the plate
and the loading.
Two S/A pairs are used to maximize the controllability of the first two modes. In this simulation, it is
assumed that r 30, p 2, and Ri 1 3 i 30.
The weight R for the first two modes is chosen as
R 1
105 to emphasize the relative importance of the
first two modes. The proposed start position and the
optimal placement of two piezoelectric S/A pairs are
shown in Figure 36. The optimal position for two S/A
pairs is at the center of the plate and along the principal
axis of the upper layer of the plate 45 and symmetric
about the diagonals.
The corresponding non-dimensional central deflection
is shown in Figure 37 and the required input voltages of
the S/A pairs in Figure 38. The free vibration damps out
within 0.07 s and the symmetry in voltage requirements
of the S/A pairs are exhibited.
If the allowable input voltages for the S/A pairs are
constrained as 1 2 42V, the optimization is
performed based on the combined modified objective
function U x. The optimal location under a 8
8 mesh
is shown in Figure 39 ( 1
10 10 ). Similar to the
cantilevered plate case, the optimal location is different

Downloaded from jim.sagepub.com by kishore bondada on December 29, 2014

243

Vibration Control of Composite Plates


y

(a)

80
60
40

Actuator voltage (Volt)

Fixed end
1

5
2
3

20
0
-20
-40

1
1

-60

(1)min=-73.17

7
-80

0.01

0.02

Volt

0.03

(b)

0.04 0.05 0.06


Time(second)

0.07

0.08

0.09

0.1

Figure 38. Required input voltages, identical for both pairs


(unconstrained case).

Fixed end

y
5
2

Fixed end

1
1

Figure 36. Optimal placement of two S/A pairs (first two modes): (a)
Start position; (b) Local optimum.

3
2
1

0.8

Figure 39. Optimal placement with constrains on the input voltages


(first two modes).

0.4
0.2
0

1.5

-0.2
-0.4

1
-0.6
-0.8
-1

0.01

0.02

0.03

0.04 0.05 0.06


Time(second)

0.07

0.08

0.09

0.1

Figure 37. Non-dimensional central deflection of clamped plate


(unconstrained case).

from the case without constraint where the S/A pairs


are located such that the input voltages required is
below the constrained limit. The corresponding nondimensional central deflection at (5, 5) is shown in
Figure 40 and the input voltages of the S/A pairs in
Figure 41, indicating the reduced control effectiveness.

Nondimensional central deflection

Nondimensional central deflection

1
0.6

0.5

-0.5

-1

-1.5

0.01

0.02

0.03

0.04 0.05 0.06


Time(second)

0.07

0.08

0.09

0.1

Figure 40. Non-dimensional central deflection of clamped plate


(constrained case).

Downloaded from jim.sagepub.com by kishore bondada on December 29, 2014

244

S. T. QUEK
40

ET AL.

effectiveness when the input voltages of the S/A pairs


are constrained, is illustrated numerically.

30

Actuator voltage (Volt)

20
10

ACKNOWLEDGMENTS

0
-10

The authors appreciate the constructive comments of


the reviewers. This study is supported in part by a
research grant provided by the National University of
Singapore.

-20
-30
-40 (1)min=-40.71
-50

0.01

0.02

0.03

Volt
0.04 0.05 0.06
Time(second)

0.07

0.08

0.09

0.1

Figure 41. Required input voltages, identical for both S/A pairs
(constrained case).

The maximum input voltages are 40.71 V compared to


73.17 V for the unconstrained case.

CONCLUSIONS
The optimal placement of piezoelectric sensor/actuator (S/A) pairs for maximum damping effect of
laminated composite plate is investigated using a finite
element formulation. To select a starting layout of the
S/A pairs for the optimal search, the maximum
integrated value of the sum of normal strains is
proposed. Direct pattern search method is then used
to obtain the local optimum. Two performance indicators are used, relating to the modal controllability and
system controllability, respectively. Based on the
numerical results for a cantilevered and a clamped
square plate, it is shown that the proposed strategy of
using the maximum sum of normal strains to obtain the
start point works well with the discrete direct pattern
search method in obtaining the optimal location of
piezoelectric S/A pairs to control the vibration of the
composite plate. If the thickness of the S/A pairs is
small, the start point may be coincident with the global
optimum. This strategy of choosing a good start point is
promising for finding the optimal placement of large
systems. It is also found that the local optimum
obtained is the global optimum. For the cantilevered
composite plate, S/A pairs located close to the fixed end
are the most and at the free end the least effective for
controlling the first two modes. The S/A pairs should be
placed near the fixed end but away from the center
line for torsional vibration control. For the clamped
composite plate, the S/A pairs placed at the centers
of the half waves are the most effective for the first
two modes. For both plates, the reduced control

REFERENCES
Barboni, R., Mannini, A., Fantini, E. and Gaudenzi, P. 2000.
Optimal Placement of PZT Actuators for the Control of Beam
Dynamics, Smart Materials and Structures, 9:110120.
Crawley, E. F. and de Luis J., 1987. Use of Piezoelectric Actuators as
Elements of Intelligent Structures, AIAA Journal, 25(10):
13731385.
Cooper, L. and Steinberg, D. 1970. Introduction to Methods of
Optimization, W. B. Saunders Company, Philadelphia.
Duch, W., Grudzicnski, K. and Diercksen, G.H.F. 1998. Minimal
Distance Neural Methods, In: World Congress of Computational
Intelligence, IJCNN98 Proceedings, pp. 12991304, Anchorage,
Alaska.
Frecker, M.I. 2002. A Review of Current Research Activities in
Optimization of Smart Structure and Actuators, Modeling, Signal
Processing, and Control, In: Vittal S. Rao, (ed.), SPIEs 9th
Annual International Symposium on Smart Structures and
Materials, Paper No. 4693-14.
Gao, F., Shen, Y.P. and Li, L.X. 2000. The Optimal Design of
Piezoelectric Actuators for Plate Vibroacoustic Control Using
Genetic Algorithms with Immune Diversity, Smart Materials and
Structures, 9:485491.
Grefenstette, J. J. 1987. Incorporating Problem Specific Knowledge
into Genetic Algorithms, In: L. Davis (ed.), Genetic Algorithms
and Simulated Annealing, pp. 4260, Morgan Kaufmann
Publishers, Pitman, London.
Hamda, H. et al. 2002. Compact Unstructured Representations for
Evolutionary Design, Applied Intelligence, 16:139155.
Hook, R., and Jeeves, T.A. 1961. Direct Search Solution of
Numerical and Statistical Problems, J. of Assn. Computing
Machinery, 8:1931.
Hwang, W.S., Hwang, W. and Park, H.C. 1994. Vibration Control of
Laminated Composite Plate with Piezoelectric Sensor/Actuator:
Active and Passive Control Methods, Mechanical Systems and
Signal Processing, 8(5):571583.
Ip, K.H. and Tse, P.C. 2001. Optimal Configuration of a Piezoelectric
Patch for Vibration Control of Isotropic Rectangular Plates,
Smart Materials and Structures, 10:395403.
Kim, J. and Ko, B. 1998. Optimal Design of a Piezoelectric Smart
Structure for Noise Control, Smart Materials and Structures,
7:801808.
Liu, X., Begg, D.W. and Matravers, D.R. 1997. Optimal Topology/
Actuator Placement Design of Structures Using SA, J. of
Aerospace Engineering, 10(3):119125.
Onada, J. and Hanawa, Y. 1992. Actuator Placement Optimization
by Genetic and Improved Simulated Annealing Algorithms,
AIAA Journal, 31(6):11671169.
Main, J.A., Garcia, E. and Howard, D. 1994. Optimal Placement and
Sizing of Paired Piezoactuators in Beams and Plates, Smart
Materials and Structures, 3:373381.

Downloaded from jim.sagepub.com by kishore bondada on December 29, 2014

Vibration Control of Composite Plates


Padula, S.L. and Kincaid, R.K. 1999. Optimization Strategies for
Sensors and Actuators Placement, National Aeronautics and
Space Administration, NASA/TM-1999-209126, Langley Research,
Langley Virginia 23681, USA.
Papalambros, P.Y. and Wilde, D.J. 2000. Principles of Optimal Design:
Modeling and Computation, 2nd Edn, Cambridge University Press,
New York, NY.
Ryall, T.G. 2001. Two Considerations for the Design of a Robust
Optimal Smart Structure where Control Energy is Expensive,
In: Smart Structures and Devices, Proceedings of SPIE,
4235:355362.
Schmidhuber, J. and Hochreiter, S. 1996. Guessing can Outperform
Many Long Time Lag Algorithms, Technical Note IDSIA19-96.

245

Suh, C.H. and Radcliffe, C.W. 1978. Kinematics and Mechanisms


Design, John Wiley & Sons, New York, NY.
Sun, D.C., Wu, H.X. and Wang D.J. 1998. Modal Control of Smart
Plates and Optimal Placement of Piezoelectric Actuators, In:
Proc. Inter. Conf. Vib. Eng., pp. 551555, Northeastern University
Press, Dalian, China.
Trosset, M.W. 2001. What is Simulated Annealing, Optimization
and Engineering, 2:201213.
Wang, S.Y., Quek, S.T. and Ang, K.K. 2001. Vibration Control of
Smart Piezoelectric Composite Plates, Smart Materials and
Structures, 10:637644.
Whitley, D., Mathias, K. and Fitzhorn, P. 1991. Delta Coding: An
Iterative Search Strategy for Genetic Algorithms, In: Proc. ICGA
4, pp. 7784.

Downloaded from jim.sagepub.com by kishore bondada on December 29, 2014

You might also like