INSTITUTE OF PHYSICS PUBLISHING SMART MATERIALS AND STRUCTURES

Smart Mater. Struct. 13 (2004) 220–226 PII: S0964-1726(04)72568-9

The application of genetic algorithms for

shape control with piezoelectric
patches—an experimental comparison
S da Mota Silva1 , R Ribeiro2 , J Dias Rodrigues2,3 , M A P Vaz2 and
J M Monteiro2
1
European Laboratory for Particle Physics (CERN), Geneva, Switzerland
2
Faculdade de Engenharia da Universidade do Porto, DEMEGI, Portugal

E-mail: jdr@fe.up.pt

Received 14 January 2003, in final form 24 November 2003

Published 7 January 2004
Online at stacks.iop.org/SMS/13/220 (DOI: 10.1088/0964-1726/13/1/026)
Abstract
The problem of shape control and correction of small displacements in
composite structures using piezoelectric actuators glued or embedded was
addressed. A finite element model based on Mindlin plate theory was used
to characterize the behaviour of the structure and the one of the actuators.
Emphasis was put on the development of an efficient and general
methodology, based on genetic algorithms, for the determination of the
optimal actuation voltages needed to apply to the piezoelectric actuators in
such a way that a pre-defined shape of the structure is achieved. The model
was investigated numerically and verified experimentally. Measurements
were carried out using electronic speckle pattern interferometry. Good
agreement was found between simulation results and the optically measured
values.
(Some figures in this article are in colour only in the electronic version)

1. Introduction The linear theory of piezoelectricity assumes that the

The modification and the correction of the shape of mechanical
structures has found an enormous range of applications in
recent years. Most of them are related either with the
study of flexible aircraft skins for improving aerodynamic
performance, or with the correction of the shape and
curvature of mirrors/antennas for high pointing accuracy
or with the compensation for temperature/humidity induced
displacements in structures with high demanding stability
requirements [1]. The technology involves the use of ‘smart’
actuators, either glued or incorporated within the structure,
optimally located and able to generate enough forces to induce
deformations according to pre-defined requirements. One of
the main candidates for distributed actuation and control, in
engineering applications such as shape control and vibration
suppression, are piezoelectric materials, such as lead zirconate
titanate (PZT), with its mechanical and electrical coupled
properties.
3 Author to whom any correspondence should be addressed.
deformations are infinitesimal and the electric field is quasi-
static, and consequently it can be described only by the
gradient of the electric potential. Tiersten [2], based on the
energy formulation, derived the linear constitutive equations
for piezoelectric materials. The issues of modelling composite
plates with integrated piezoelectric materials have also been
addressed. Hwang and Park [3] used the classical laminated
plate theory and Miller and Abramovich [4] the first-order
shear deformation theory. However, for the prediction of
stresses of highly anisotropic plates and/or moderately thick
composite plates these theories may lead to significant errors.
More accurate mathematical models can then be used: either
3D and hybrid models [5] or models based on high order
displacement fields [6].
Koconis et al [7, 8] presented one of the first studies on
the shape control of structural elements. Determination of
the shape due to specified voltages and determination of the
voltages required to produce pre-defined shapes on composite
beams, plates and shells were reported. Irschik et al [9]

0964-1726/04/010220+07$30.00 © 2004 IOP Publishing Ltd Printed in the UK 220

 The application of GAs for shape control with piezoelectric patches—an experimental comparison
developed an exact solution for the static shape control of
adaptive structures using piezoelectric actuation. Banks
et al [10] reported experimental results on a Euler–Bernoulli
cantilever beam.
Besides establishing an accurate mathematical model able
to deal with shape control applications, one important issue
that needs to be addressed is the determination of the optimal
location of the piezoelectric actuators together with the optimal
actuation voltages [11]. A proper placement of the actuators
is a critical factor for the success and performance of the
structure. Several optimization techniques are available for
the purpose. Chandrashekhara and Varadarajan [12] used the
Lagrange multiplier approach and a quadratic cost function of
nodal deflection errors for determining the optimal actuator
voltages. Optimization techniques such as genetic algorithms
(GAs) and simulated annealing (SA) have also been used to
determine optimal actuator locations in truss structures for the
correction of static deformations. Kincaid [13] applied the SA
algorithm to the static distortion minimization problem. Onoda
and Hanawa [14] applied a modified GA and the improved
simulated annealing (ISA) algorithm to the optimization of
actuators placement in truss structures for static distortion
correction. Cook and Crossley [15] presented a genetic
algorithm approach to place discrete actuators on an aircraft
wing to produce control moments about all three axes.
In literature, the great majority of the studies reported
so far on the shape control of structures with piezoelectrics
are numerical, and only a few of them consider experimental
validation. The present work focuses on the experimental
validation of a straightforward and general methodology for
determining the optimal actuation voltages needed to apply to
piezoceramic actuators for changing the shape of a structure or
for correction of static deformations. A finite element method,
based on the Mindlin theory of plates, was used. Genetic
algorithms were used as the optimization technique and the
optimization function was defined, in terms of the surface
error, by the square root of the deviation between the pre-
defined shape and the achieved shape obtained when applying
the optimum actuation voltages.
2. The optimization method: genetic algorithms
In the very beginning the most common optimization
techniques relied on graphical methods. These methods
were in general very simple and straightforward; however,
important limitations were related to the number of variables
it was possible to deal with. As an alternative, numerical
optimization methods were developed over the last four
decades. Conventional optimization search methods can be
divided in three groups: algorithms that make use of gradients,
enumerative schemes and random methods [16].
The algorithms based on gradients assume the continuity
of the search space, which in real world problems is seldom
the case. Over the broad spectrum of optimization functions,
these methods show lack of robustness and are local in scope.
Enumerative schemes are based on a straightforward idea:
within a finite search space, or discretized infinite search
space, the algorithm looks at the objective function values
at every point in space, one at a time. Although simple
and obvious, these schemes suffer from lack of efficiency,
while presenting an huge computational task. Random search
algorithms perform random walks on the problem space and
record the best optimum values discovered so far. These
methods do not use any knowledge from previous results and
hence fail the efficiency requirement. Genetic algorithms
make use of randomized techniques that, contrary to the later
methods, are not directionless, since the knowledge from
previous incursions to the search space is considered.
2.1. Main concepts
One of the most powerful features of GAs is their simplicity.
Several aspects make them very attractive and a powerful
alternative over other techniques. Namely, GAs work, in
general, with a coding of the parameter set, not with the
parameters themselves; GAs search from a population of points
or candidate solutions, not a single point; GAs use the function
to be optimized (fitness function), not derivatives or other
auxiliary knowledge of that function. As a result of the use
of randomization techniques, GAs use probabilistic transition
rules, not deterministic ones. Due to its stochastic nature, only
an estimate of the exact optimum can be found: GAs will only
by chance find the exact optimum, whereas traditional methods
will find it exactly. In this case, the question is to know if the
found solution is close enough, which is generally somehow
arbitrary and dependent on the application.
2.2. Main genetic operators
Conceptually, GAs are iterative procedures which maintain a
population of candidate solutions to the problem of interest,
the so-called chromosomes. Each chromosome is constituted
by a number of individual structures called genes. Each
gene is associated with a specific parameter/variable of the
search space. Chromosomes are usually constant-length
sequences, and traditionally binary numbers; however, in
practice they can be anything including a mix of integers
and real numbers, or a mix of numbers and character strings.
During each iteration step, called generation, all individuals
or solutions in the current population are assigned a fitness
value, based upon some analysis using the decoded values
of each individual’s design variables. On the basis of those
evaluations, a new population of candidate solutions is formed
for use in further operations. During the selection procedure,
highly fit individuals are more likely to be selected than unfit
members. The idea is to allow more fitted candidate solutions
to survive and procreate. The overall process is responsible for
the speed of evolution and is therefore one of the key features
for the success of an evolutionary algorithm. The selection
process can be implemented in the algorithm in a number of
ways, of which the tournament selection procedure was the one
selected. Based on the idea of ranking, this procedure selects
randomly, in a single iteration, some number k of individuals
from the population. The best individual from this group is
then selected as a parent for the next generation. The process
is repeated until the number of individuals in the breeding
population is equal in size to the number of candidate solutions
in the initial population. A consequence of the process is that
the average fitness of the breeding or parent’s population will
be higher than the fitness values of the preceding generations.
221
S da Mota Silva et al
Figure 1. The general layout of a genetic algorithm’s
implementation code.
After selection, a crossover genetic operator is applied
to two (or more) chromosomes, called parents, according
to a pre-defined crossover probability. This operator allows
the recombination and exchange of portions of genetic
material resulting in new chromosomes called offsprings.
Although crossover may take different forms, it is generally
implemented by choosing one, or more, random cut-point, the
crossover point, and exchanging the segments from this point
onwards.
The last operator involved in the process is the mutation
operator. It is a background operator that produces, according
to a pre-defined mutation probability, a random alteration of
a gene restoring genetic diversity. This operator is generally
responsible for providing the genes that were not present in
the initial population, and replacing the genes lost from the
population during the selection process so they can be tried in
a new context. In practice, the risk of the search ending at a
local optimum is decreased.
The goal of the process, which is well suited to complex,
discontinuous and/or multimodal search spaces, is to select and
combine features associated with good performance in order
to identify a nearly global optimum solution (see figure 1).
2.3. Control parameters
The main parameters controlling the search behaviour are:
the population size, which is able to affect both the ultimate
performance and the efficiency of the algorithm; the crossover
probability, which controls, as already mentioned, the
frequency at which the crossover operator is applied; and
the mutation probability, whose main effect is to increase
the population variability and the selection strategy. The
influence of these parameters on the algorithm’s behaviour may
be described in terms of two primary factors: selective pressure
and population diversity. As selective pressure increases, the
search focuses on highly fitted members of the population
often excluding low fitted strings from the whole process.
Population diversity is related with the distribution throughout
the search space of strings in the breeding population. The
relation and a correct balance between these two opposite
factors is a key to a successful global optimization search [17].
Several methods are available to control the selective pressure
and population diversity and to improve the genetic search.
One of the most common is an elitist strategy.
An elitist strategy can be employed to avoid the situation
that the best solution in the population disappears due
to sampling errors, crossover or mutation. This strategy
guarantees that the best performance survives intact to the next
222
generation by comparing the best solutions of two consecutive
generations. In the case that the best solution of the previous
generation is better than the one from the current generation,
the former will replace the worst solution of the current
generation. Occasionally, this method may result in a search
excessively focused on the best solution found, at the expense
of not undertaking a wider search for other potentially good
solutions [17].
Genetic algorithms are sufficiently robust to allow
several configuration alternatives; however, performance and
convergence characteristics may benefit from selecting one
configuration over another for a given problem. A judicious
choice of the objective function representing the optimization
problem to be solved, an appropriate choice of the genetic
operators, as well as the different control parameters, play an
enormous role in the development of an efficient multimodal
optimization algorithm.
3. The mathematical model
Hamilton’s principle was used to derive the equations of motion
for a laminated composite plate containing piezoelectric
patches. For coupled electromechanical systems this principle
states that
t2
δ(T − U + Wext ) dt = 0 (1)
t1
where t1 and t2 are arbitrary instants, T is the kinetic energy, U
the potential energy (including strain and electrical energies)
and Wext the work done by external forces. For coupled electro-
mechanical systems, the kinetic energy, the potential energy,
and the work done by discrete applied external forces are,
respectively, defined as


T = 1
2
{q̇}T {q̇} dV,
V


U= 1
2
[{S}T {T } − {E}T {D}] dV, (2)
V

nf
Wext = {q}T {Fc }
i=1
where {q̇} is the velocity vector,  the mass density, {S} the
strain vector, {T } the stress vector, {E} the electrical field
vector, {D} the vector of electrical displacements, {Fc } the
external force vector acting at (xi , yi ), n f the number of
applied external forces, and V the volume of the structure.
3.1. Constitutive equations
The linear piezoelectric constitutive equations coupling the
elastic field and the electric field vector can be expressed by the
direct and the converse piezoelectric equations respectively:
{D} = [e]{S} + [ s ]{E} {T } = [C E ]{S} − [e]T {E} (3)
with [e] being the piezoelectric module matrix, [ s ] the
dielectric constants matrix evaluated at constant strain, and
[C E ] the matrix of the elastic coefficients for the piezoelectric
material at constant electric field. Piezoceramic materials can
be generally considered as transversally isotropic in the plane
normal to the polarization axis, z. In this case, equation (3)
can be considerably simplified.
 The application of GAs for shape control with piezoelectric patches—an experimental comparison
3.2. Finite element discretization
Under the same assumptions and restrictions as in the classical
laminate theory, the displacements u, v and w at a point
(x, y, z) from the median surface of the plate are expressed
by the mid-plane displacements u 0 , v0 and w0 and by the
rotations θx and θ y of a transverse normal about the x and
y axes respectively, as
u(x, y, z, t) = u 0 (x, y, t) + zθ y (x, y, t)
v(x, y, z, t) = v0 (x, y, t) − zθx (x, y, t)
w(x, y, z, t) = w0 (x, y, t).
Substituting equation (4) into the infinitesimal strain relations
one obtains {S} = {S 0 } + z{κ} + {S s }, where {S 0 } are
the membrane strains, {κ} the bending strains, and {S s } the
transverse shear strains. The strain relations are not functions
of the material, and consequently the same expressions are
valid in any point of the structure. The generalized strain
vector, {S ∗ }, can be defined in terms of the generalized
displacement vector {q} and the differential operator matrix
[L] by {S ∗ } = [L]{q}.
A piezoelectric material, either embedded or glued into the
structure, is able to induce a strain when an electrical voltage
is applied to it. This induced strain is transmitted to the host
structure by the bonding layer between them. A four-node
element based on the Mindlin theory of plates (Mindlin 1951)
with six degrees of freedom in each node and one electrical
degree of freedom per layer was used in the model [18]. The
discrete equations of motion can be derived by defining the
displacement and the electrical potential in terms of i nodal
variables via the shape functions matrices [Nq ] and [Nφ ]:
{q} = [Nq ]{qi } {φ} = [Nφ ]{φi }
where {qi } are the mechanical and {φi } the electrical
generalized coordinates.
3.3. Equations of motion
The equations of motion of the electro-mechanical system
can be obtained after substituting all the energy terms in
equation (1) and by considering equation (5). Allowing
arbitrary variations of {q} and {φ}, two matrix equations in
the generalized coordinates are obtained:
[Mmm ]{q}
¨ + [K mm ]{q} − [K em ]{φ} = F
[K em ]T {q} + [K ee ]{φ} = 0
where the mass and the mechanical stiffness matrix are defined
as

and accurately detected with optical methods like the electronic
[Mmm ] = [Nq ]T [Nq ] dV
V

(8)
[K mm ] = T E
[Bq ] [C ][Bq ] dV
V presented in figure 2.
and the coupled electrical/mechanical stiffness, the piezoelec-
tric stiffness matrix, and the forcing matrix, respectively, as


[K em ] = [Bq ]T [e][Bφ ] dV
V


[K ee ] = [Bφ ]T [ s ][Bφ ] dV (9)
V

nf
{F} = [Nq ]T {Fc }.
i=1
(4)
3.4. Shape control applications
Equations (6) and (7) represent, respectively, the actuator and
the sensor equations used in the simulation models. The
first equation can be used to solve shape control applications.
Namely, shape control problems can be considered quasi-static
if slow time varying disturbances are assumed. For this case the
nodal displacements, {q}, can be easily computed, considering
equation (6) in a free load condition, as a function of the
stiffness of the structure and of the applied voltages, {φ}, by
[K mm ]{q} − [K em ]{φ} = 0 ⇔ {q} = [K mm ]−1 [K em ]{φ}. (10)
When considering plate elements, the shape of a structure is
mainly described by the shape of its mid-plane, which itself is
described by the transverse displacement of the finite element
mesh nodes. Therefore, the deviation between the pre-defined
displacement field function and the achieved displacement can
be defined as the sum of the deviations at the n nodal points,
and the objective function, J , is then given by

n
J2 = (γi − qi )2 (11)
i=1
where γi is the pre-defined displacement at the i th node, and
(5) qi the transverse displacement at the same node. This function
gives an estimate of the variance of the current displacement
values with respect to the pre-defined ones.
The shape control problem, or the correction of static
deformations, of a given structure with a given layout of
piezoelectric actuators consists of finding a set of actuator
voltages, φi , that minimizes the error function J and satisfies
the constraint φmin
φi
φmax , where φi is the actuation
voltage of the i th actuator, and φmin and φmax the lower and
upper saturation voltages of the actuators. This represents the
minimization problem—fitness function and constraints—to
be solved by genetic algorithms. The use of this optimization
technique will enable, in the future, the extension of the present
formulation in such a way that both optimal voltages and
(6) optimal placement of the piezoelectric actuators in the structure
can be determined (discontinuous problem).
(7)
4. Applications
Deformations in the micron range and over a surface are easily
speckle pattern interferometry (ESPI) [19]. This technique,
capable of providing three-dimensional displacements and
dynamic response characteristics of surfaces, is schematically
223
S da Mota Silva et al

lens
Beam
Splitter Object
Shutter
lens

LASER Reference Beam

Monitor
lens Figure 4. The phase map resulting from the captured raw fringe
CCD pattern for case 1.

Figure 2. A schematic representation of ESPI.

30
3 3

4
12

3 48

24 2
12

Figure 5. The phase map resulting from the captured raw fringe
Y-axis 1 24
pattern for case 2.
X-axis 153

Figure 3. The aluminium plate used in the experiment (mm).

The deformation of a plate in two pre-defined directions 6.63

according to the actuation voltages obtained by GAs will be 5.25
experimentally verified using this method. The experimental 3.87
set-up comprises a clamped free aluminium plate, 48 mm Z 2.49
wide, 153 mm long and 0.5 mm thick, with four bonded 1.10
piezoceramics. Figure 3 presents the layout. – 0.28
152.67
64.36
The piezoceramic elements (PX5-N from Philips 48.18 122.20
Components) are 0.3 mm thick, 24 mm long and 12 mm wide, 32.00 91.73
Y 15.82 61.27 X
with the electro-mechanical properties shown in table 1. – 0.36 30.80
Table 2 shows the considered pre-defined shape and the
actuation voltages, obtained by GAs, needed to apply to each Figure 6. A representation of the displacements for case 1.
of the four piezoelectric elements in order to achieve that same
shape. The plate is intended to have a combined movement of
bending and torsion.
The voltages in each piezoelectric actuator were limited
to −10 and +20 V. A finite element mesh of 4 × 10 elements 13.14
was assumed. The optimization method, based on genetic 10.40
7.65
algorithms, made use of a finite element analysis to model
Z 4.90
the structure. The selection scheme was tournament selection
2.16
(k = 2), and an elitist strategy was adopted. Each chromosome
– 0.99
has a length equal to the number of piezoelectric actuators in 64.36 152.67
48.18 122.20
the structure, and real coding is used. Single-point and uniform 32.00 91.73
crossover were implemented together with uniform mutation. Y 15.82 61.27 X
– 0.36 30.80
A population size of 25 individuals, a probability of crossover
equal to 0.75, a probability of mutation equal to 0.20, and a Figure 7. A representation of the displacements for case 2.
maximum number of generations of 50 000 were considered
in the calculations. Multiple runs were conducted.
The actuation voltages applied to each piezoelectric The inclination of the fringe pattern in both figures implies the
element, with a precision of ±0.1 V, generated stable existence, in practice, of a small rotation along the longitudinal
deformations with time. Figures 4 and 5 show the phase axis of the plate (torsion effect) as already expected. The
map of the raw fringe pattern captured for case 1 and case 2, displacement field of the plate for both cases is shown in
respectively. figures 6 and 7. A localized modification of the displacement
It is worthwhile mentioning that the interferometer is field in the region of the piezoelectric elements is visible.
sensitive to displacements normal to the plate’s surface and The pre-defined displacement field, simulations, and
that each fringe corresponds to a region of equal displacements. experimental measurements are compared for case 1 and case 2

224
 The application of GAs for shape control with piezoelectric patches—an experimental comparison

Table 1. Material properties for the PX5-N piezoceramic material (data from the manufacturer).
E
C11 (N m−2 ) 13.11 × 1010 E
C44 (N m−2 ) 2.564 × 1010 d15 (m V−1 ) 515 × 10−12
E
C12 (N m−2 ) 7.984 × 1010 C66 (N m−2 )
E
2.564 × 1010 11
t
/0 1800
E
C13 (N m−2 ) 8.439 × 1010 d31 (m V−1 ) −215 × 10−12 33
t
/0 2100
E
C33 (N m−2 ) 12.31 × 1010 d33 (m V−1 ) 500 × 10−12  (kg m−3 ) 7800

Table 2. Pre-defined displacement fields (m) and optimal actuation voltages (V).
Displacement field Vpiezo1 Vpiezo2 Vpiezo3 Vpiezo4
−4
Case 1 γ (x, y) = (1.91x + 0.88x y + 0.19x) × 10
2
−1.7 −2.5 4.9 5.0
Case 2 γ (x, y) = (3.83x 2 + 1.76x y + 0.38x) × 10−4 −3.4 −5.0 9.8 10.0

Figure 8. Pre-defined shape, simulations and experimental Figure 10. Pre-defined shape, simulation and experimental
measurements for y = 24 mm (case 1). measurements for x = 77 mm (case 1).

Figure 9. Pre-defined shape, simulations and experimental Figure 11. Pre-defined shape, simulation and experimental
measurements for y = 24 mm (case 2). measurements for x = 77 mm (case 2).

in figures 8 and 9 as functions of the longitudinal coordinate
at points y = 24 mm.
It was observed that for case 2, figure 9, in the region
near the free edge of the plate, the fringes present an area
where their space frequency is too close to the upper limit
of the measurements for that reference. This introduces
some errors in the determination of the displacements for
that area and explains the loss of linearity and information
of the displacements in the extremity of the plate. Using
the available information of the displacement field along the
middle longitudinal axis of the plate, the tip displacement was
extrapolated for this case. A value of 14.43 µm was obtained,
representing a difference of 3% from the simulated value and
a difference of 4.1% from the pre-defined one. In general, an
excellent agreement is found between the pre-defined shape,
simulations and experimental measurements obtained with
ESPI along the longitudinal coordinate.
The pre-defined displacement field, simulations and
experimental measurements are compared in figures 10, 11
and figures 12, 13 as functions of the transverse coordinate
along x = 77 and 153 mm, respectively.
For case 1, figure 10, it can be seen that the maximum
error between the simulation results and the pre-defined curve
is equal to 15% at one of the extremities of the plate (y/ W =
1.0), while between the experimental results and the pre-
defined ones this value equals 3%. The comparison for case 2,
figure 11, is less satisfactory. Please note the almost constant
offset along the width between measurements and the pre-

225
S da Mota Silva et al
Figure 12. Pre-defined shape, simulation and experimental
measurements for x = 153 mm (case 1).
Figure 13. Pre-defined shape, simulation and experimental
measurements for x = 153 mm (case 2).
defined curve. Errors attain, once again at the extremity of
the plate and for both cases, a value of the order of 15%.
For case 1, figure 12, it is still verified that the maximum
differences between simulation, experimental and pre-defined
results occur at one of the free edges of the plate (y/ W =
1.0). Namely, a maximum error of 14% is found between
the experimental results and pre-defined ones, while between
simulation results and pre-defined ones this value drops to 3%.
For case 2, figure 13, the experimental measurements made
along x = 153 mm present a loss of linearity. This fact, as
already explained, is related to the introduction of considerably
high measurement errors that invalidate any comparison. On
the other hand, the maximum error between the simulation
results and the pre-defined curve is equal to 3%, and an
excellent agreement is obtained.
5. Conclusions
A systematic and general methodology, using a finite element
code and genetic algorithms, for the shape control and/or
correction of static deformations of adaptive structures, was
proposed and verified experimentally. The square root of
the error between the nodal pre-defined displacement and the
achieved displacement was considered as the performance
index. Throughout the presented results only glued
piezoelectric elements are considered; however, the developed
methodology also supports embedded ones. Maximum
errors of 15% were obtained between simulation/experimental
results and the pre-defined curve. The practicality and the
226
effectiveness of using genetic algorithms to determine the
optimal actuation voltages was demonstrated and the adopted
methodology validated.
References
[1] Ribeiro R, da Mota Silva S, Dias Rodrigues J and Vaz M 2000
Genetic algorithms for optimal design and control of
adaptive structures Proc. SPIE 3984 268–78
[2] Tiersten H F 1969 Linear Piezoelectric Plate Vibrations (New
York: Plenum)
[3] Hwang W and Park H 1993 Finite element model of
piezoelectric sensors and actuators AIAA J. 31 930–7
[4] Miller S E and Abramovich H 1995 A self-sensing
piezolaminated actuator model for shells using a first order
shear deformation theory J. Intell. Mater. Struct. 6 624–38
[5] Kapuria S, Sengupta S and Dumri P C 1997
Three-dimensional solution for simply-supported
piezoelectric cylindrical shell for axisymmetric load
Comput. Methods Appl. Mech. Eng. 140 139–55
[6] Correia V, Soares C M and Soares C M 1999 A numerical
model for the optimal design of composite laminated
structures with piezoelectric laminae Proc. SPIE 3667
427–37
[7] Koconis D, Kollar L and Springer G 1994 Shape control of
composite plates and shells with embedded actuators I.
Voltages specified J. Compos. Mater. 28 415–58
[8] Koconis D, Kollar L and Springer G 1994 Shape control of
composite plates and shells with embedded actuators II.
Desired shape specified J. Compos. Mater. 28 262–85
[9] Irschik H, Adam C, Heuer F and Ziegler F 1998 An exact
solution for static shape control using piezoelectric
actuation adaptive structures ASME Publication 15 103–13
[10] Banks H, Ito K and Wang Y 1993 Computational methods for
identifications and feedback control in structures with
piezoceramic actuators and sensors J. Intell. Mater. Struct. 4
469–76
[11] da Mota Silva S, Ribeiro R, Dias Rodrigues J and Vaz M 2001
Shape control of structures with PZT actuators using
genetic algorithms Proc. ICCM-13: 13th Int. Conf. on
Composite Materials (Beijing, June 2001) CD-ROM
[12] Chandrashekhara K and Varadarajan S 1997 Adaptive shape
control of composite beams with piezoelectric actuators
J. Intell. Mater. Struct. 8 112–24
[13] Kincaid 1991 Minimization distortion in truss structures: a
comparison of simulated annealing and tabu search 32nd
AIAA/ASME/ASCE/AHS/ASC Structures, Structural
Dynamics and Materials Conf. (Baltimore, MD, April 1991)
AIAA Paper 91-1095
[14] Onoda J and Hanawa Y 1992 Optimal locations of actuators
for statistical static shape control of large space structures:
a comparison of approaches 33rd
AIAA/ASME/ASCE/AHS/ASC Structures, Structural
Dynamics and Materials Conf. (Dallax, TX, April 1992)
AIAA Paper 92-2558
[15] Cook A and Crossley W 2001 Investigation of genetic
algorithm approaches for smart actuator placement for
aircraft maneuvering 39th AIAA Aerospace Sciences
Meeting (Reno, NV, January 2001) AIAA Paper 2001-0924
[16] Goldberg D E 1989 Genetic Algorithms in Search,
Optimization, and Machine Learning (New York:
Addison-Wesley)
[17] Simpson M and Hansen C 1996 Use of genetic algorithms to
optimize vibration actuator placement for active control of
harmonic interior noise in a cylinder with floor structure
Noise Control Eng. J. 44 169–84
[18] Suleman A and Venkayya V B 1995 A simple finite element
formulation for a laminated composite plate with
piezoelectric layers J. Intell. Mater. Struct. 6 776–82
[19] Silva Gomes J, Monteiro J and Vaz M 2000 NDI of interfaces
in coating systems using digital interferometry Mech.
Mater. 32 837–43