Shape Control of Structures with PZT Actuators Using

Genetic Algorithms
S. da Mota Silva 1;2 , R. Ribeiro 3  , J. Dias Rodrigues 2 , M. Vaz 2
1European Laboratory for Particle Physics, CERN - CH 1211 Geneva 23 Switzerland
2 Faculdade de Engenharia da Universidade do Porto, Departamento de Mecânica, Porto, Portugal
3 Universidade do Minho, Departamento de Fisica, Guimarães, Portugal

SUMMARY: The use of piezoelectric actuators for shape control and correction of static deformations
is considered. Genetic algorithms (GA) are implemented to calculate the optimal layout and the optimal
voltages, applied to the piezoelectric actuators glued in the structure, which minimise the error between
the achieved and the pre-defined shape. The need of an optimisation algorithm able to solve efficiently
this problem without imposing a large number of requirements and restrictions lead us to use these algo-
rithms. Simulations and experimental measurements are compared for different geometries with different
boundary conditions and pre-defined displacement fields. Experimental results consider only beam and
plate structures made of isotropic materials, however the developed methodology is general and can be
applied to a more complex geometry made of orthotropic materials.

KEYWORDS: shape control, genetic algorithms, actuator layout, optimum design, PZT actuators,
adaptive structures.

INTRODUCTION
Shape control of flexible structures can improve the aerodynamic performance of lifting surfaces, correct
the shape of antennas or mirrors as well as compensate quasi-static displacements due to temperature or
humidity variations. The most common methods to control the shape of a structure consists in embed-
ding piezoelectric materials in it. Crawley and De Luis (1987)1 developed static and dynamic analytical
models for structures with distributed sensors and actuators, glued or embedded in the structure. These
models are able to predict, a priori, the response of the structure. Koconis, Kollar and Springer (1994)2 ; 3
addressed the shape control problem: calculation of the shape of the structure due to specified voltages
and calculation of the required voltages to produced a pre-defined shape. Chandrashekhara and Varadara-
jan (1997)4 used the Lagrange multiplier approach to determine the optimal actuator voltages needed to
attain the pre-defined shape of a composite beam.
This paper proposes the use of a straightforward method, based on finite element analysis and GA, to
calculate the optimal position and the optimal actuation voltages of the piezoceramic (PZT) elements
bonded on a structure in order to induce that same structure to deform in a certain pre-defined way.
The only constraints are the saturation voltages of the piezoelectric elements and the maximum number
of actuators allowed within the structure. Simulation and experimental measurements validate the all
procedure.

MATHEMATICAL MODEL AND OPTIMISATION ALGORITHM

Shape control problems can be considered quasi-static in the presence of slow time varying disturbances.
In this case the displacement field can be easily computed as a function of the stiffness of the structure
and of the applied actuation voltages, fg, by5 ; 6

fqg = [Kmm ];1 [Kem ]fg (1)

 On leave of absence from the European Laboratory for Particle Physics

1
The determination of the mechanical stiffness, [Kmm ], and the coupled electrical/mechanical stiffness
matrices, [Kem ], is based in the Mindlin theory of plates. The equations of motion were discretised
in four-node elements with 6 degrees of freedom per node plus an electrical degree of freedom per
piezoelectric layer.
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. There-
fore, the error between the pre-defined displacement field function and the achieved displacement field,
can be defined as the sum of the errors at the n nodal points, and the fitness or objective function, J , is

X ;q
then given by
n
J 2
= ( i i )2 (2)
i=1
where i is the pre-defined displacement at the i ; th node and qi the transverse displacement at the
same node. This objective function gives an estimate of the variance of the values with respect to the
pre-defined displacement. The shape control problem, or the correction of static deformations, of a given
structure with a given layout of piezoelectric actuators, consists, therefore, in finding a set of actuator
voltages, i , that minimises the error between the desired shape function and the achieved one (Equation
2) subjected to the following constraint

min  i  max (3)

where i is the actuation voltage of the i ; th actuator and min and max the lower and upper satura-
tion voltages. Equations 2 and 3 represent the minimisation formulation problem - fitness function and
constraints - to be solved by genetic algorithms.
These algorithms are stochastic optimisation techniques based in the natural evolutionary process and
on the survival-of-the-fittest7 ; 8 . They are iterative procedures which maintain a population of candidate
solutions of the problem called chromosomes. Each chromosome is constituted by a number of individual
structures called genes. During each iteration step, generation, each solution is evaluated to give some
measure of its fitness, and, on the basis of those evaluations a new population of candidate solutions is
formed. Then crossover and mutation operators are applied to this population allowing recombination
and exchange of genetic material (Fig. 1).

Figure 1. General structure of genetic algorithms.

Crossover is the main genetic operator under which two chromosomes, called parents, combine portions
of their internal representation generating new chromosomes called offsprings. The crossover probabil-
ity, pc , associated to this operator, is defined as the ratio of the number of offspring produced in each
generation to the population size. Mutation is a background operator that produces a random alteration
of a gene. The mutation probability, pm , gives the expected number of mutated genes, and controls the
rate at which new genes are introduced in the population for trial.

2
The process by which strings are copied to the new population according to their objective function
values or fitness is called reproduction or selection. The new solutions are the result of a random selection
procedure that ensures, at each generation, that the ”good” solutions reproduce, while the relatively ”bad”
ones die. An elitist strategy can be employed to avoid that the best solution in the population disappears
due to sampling errors, crossover or mutation. The reproduction operator can be implemented in the
algorithm in several ways, of which, the tournament selection procedure is one of the most commonly
used and the one adopted. This procedure belongs to the fitness-proportional selection methods and is
based on the probability distribution of the fitness values.9

RESULTS AND DISCUSSION

The procedure to find the optimal voltages of a structure in order to modify its shape was experimentally
verified. The first experimental set-up comprises 2 different structures: an aluminium beam (clamped-
free and clamped-clamped) and an aluminium plate (clamped-free). Then, the algorithm was slightly
modified in order to accommodate also the determination of the optimal localisation of the piezoelectric
actuators.

Experimental results: aluminium beam with 3 PZT

This experimental set-up comprises an aluminium beam with three bonded piezoceramics. Figure 2
presents schematically the layout of the beam for two different boundary conditions: clamped-free and
clamped-clamped.

Figure 2. Layout of the aluminium beam, 1.0 mm thick, together with the piezoelectric elements, 0.3 mm
thick, used in the experiment. Units are in mm.

The electro-mechanical properties given by the manufacturer of the piezoceramic elements (PX5-N from
Philips Components) are presented in Table 1. The saturation voltage was set to 150 Volt.
Displacement measurements were done by a non-contact capacitive displacement sensor (MC900 Fogale
nanotech). With a 5 mm measurement range the sensor had a maximum resolution of 0.5 m at 400Hz.

Clamped-free aluminium beam

Considering the clamped-free geometry two different displacement fields were arbitrarily chosen. Bound-
ary conditions and magnitude of displacements compatible with the dimensions of the structures were
the only two criteria to be fulfil. The actuation voltages applied to each of the 3 piezoelectric elements
in order to match the pre-defined displacement fields were calculated by GA. A population size of 25

3
Table 1. Material properties for the PX5-N piezoceramic material (data from the manufacturer). [CE ]
is the matrix of elastic coefficients at constant electric field, [d] the piezoelectric module matrix, [s ] the
dielectric constants matrix evaluated at constant stress and % the material density.

C11E [N=m2 ] 13.11x1010 d31 [m=V ] -215x10;12

C12E [N=m2 ] 7.984x1010 d33 [m=V ] 500x10;12
C13E [N=m2 ] 8.439x1010 d15 [m=V ] 515x10;12
C33E [N=m2 ] 12.31x1010 t11 =0 1800
C44E [N=m2 ] 2.564x1010 t33 =0 2100
C66E [N=m2 ] 2.564x1010 % [Kg=m3 ] 7800

individuals, a probability of crossover equal to 0.85, a probability of mutation equal to 0.15 and a maxi-
mum number of generations of 15000 were considered in the calculations. Table 2 shows the pre-defined
displacement fields and the actuation voltages obtained by the optimisation algorithm.

Table 2. Pre-defined displacement field and actuation voltages obtained by GA (clamped-free aluminium
beam). L is the length of the beam and x the longitudinal coordinate.

Displacement Field Vpiezo1[V ] Vpiezo2 [V] Vpiezo3 [V]

Case 1 (x) = 0:0002( Lx )2 17 36 26
Case 2 (x) = 0:0004( Lx )2 34 72 51

Figures 3 and 4 show the comparison between the two pre-defined displacement fields, simulation results
from finite element calculations and experimental measurements.
 
6LPXODWLRQ 6LPXODWLRQ
3UHB'HILQHG 3UHB'HILQHG
([SHULPHQWDO ([SHULPHQWDO
7UDQVYHUVH'LVSODFHPHQW>PP@

7UDQVYHUVH'LVSODFHPHQW>PP@

 

 

 

 
           
/RQJLWXGLQDO3RVLWLRQ>[/@ /RQJLWXGLQDO3RVLWLRQ>[/@

Figure 3. Case1:clamped-free. Comparison Figure 4. Case2:clamped-free. Comparison

between the pre-defined shape, simulation re- between the pre-defined shape, simulation re-
sults and experimental measurements. sults and experimental measurements.

In general, a very good agreement is obtained. The 2 between the pre-defined results and the sim-
ulation ones and between the pre-defined results and the experimental ones was calculated. For case
1 values of 0.0021 mm and 0.0073 mm were obtained while for case 2 one obtains 0.0042 mm and

4
0.024 mm, respectively. For both cases, experimental results are more dispersed from pre-defined values
than simulation results. For case 1, this dispersion in the experimental results is compatible with the error
associated with each measurement.

Clamped-clamped aluminium beam

The same aluminium beam was now clamped on both sides. Table 3 shows the two pre-defined displace-
ment fields, considered for this case, and the optimal actuation voltages obtained in this case. The GA
parameters were kept unchanged.

Table 3. Pre-defined displacement field and actuation voltages obtained by GA (clamped-clamped alu-
minium beam).

Displacement Field Vpiezo1[V ] Vpiezo2 [V] Vpiezo3 [V]

Case 1 (x) = 0:00004(1 ; cos( 2x
L )) -53 80 -56
Case 2 (x) = 0:000075(1 ; cos( 2x x
L ))( L ) 17 74 -121

Figures 5 and 6 show the comparison between the two pre-defined displacement fields, simulation results
and experimental measurements.
 
6LPXODWLRQ 6LPXODWLRQ
3UHB'HILQHG 3UHB'HILQHG
([SHULPHQWDO ([SHULPHQWDO
7UDQVYHUVH'LVSODFHPHQW>PP@

 
7UDQYHUVH'LVSODFHPHQW>PP@

 

 

 

 
           
/RQJLWXGLQDO3RVLWLRQ>[/@ /RQJLWXGLQDO3RVLWLRQ>[/@

Figure 5. Case1:clamped-clamped. Compari- Figure 6. Case2:clamped-clamped. Compari-

son between the pre-defined shape, simulation son between the pre-defined shape, simulation
results and experimental measurements. results and experimental measurements.

For the first case, the desired shape is achieved with relatively small voltages, with piezoelectric elements
1 and 3 presenting identical voltage values: -53 Volt and -56 Volt respectively. For the second case, which
corresponds to a non symmetry shape, the voltage level is higher in the piezoelectric element number
3, -121 Volt, and opposite in sign to the one of piezoelectric element number 1, +17 Volt. The standard
deviation between the pre-defined results and the simulation ones and between the pre-defined results and
the experimental ones was calculated. For case 1 values of 0.0018 mm and 0.0041 mm were obtained
while for case 2 one obtains 0.0024 mm and 0.0033 mm, respectively. As for the case of the clamped-free
beam, experimental results are more dispersed from pre-defined values than simulation results. However,
this dispersion in the experimental results is compatible with the error associated with each measurement
and in general a very good agreement is obtained for both cases.
The last two cases validates both the finite element code being used as well as the optimisation algorithm.

5
Experimental results: aluminium plate with 4 PZT
The shape control in two directions will be experimentally verified using a plate structure. The exper-
imental set-up comprises an aluminium plate, 0.5 mm thick, with four bonded piezoceramics (electro-
mechanical properties showed in Table 1). Figure 7 presents the layout.

Figure 7. Aluminium plate, 0.5 mm thick, used in the experiment. Units in mm.

Clamped-free aluminium plate

Table 4 shows the assumed pre-defined shape and the actuation voltages, obtained by GA, needed to
apply to each 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.

Table 4. Pre-defined displacement field and actuation voltages obtained, by GA, for the clamped-free
aluminium plate.

Displacement Field Vpiezo1 [V ] Vpiezo2 [V] Vpiezo3 [V] Vpiezo4 [V]

(x; y) = 0:0076x2 + 0:0035xy + 0:00075x -68 -100 196 200

A population size of 25 individuals, a probability of crossover equal to 0.75, a probability of mutation

equal to 0.20 and a maximum number of generations of 50,000 were considered in the calculations.
The voltages in each piezoelectric actuator were limited to -100 Volt and +200 Volt. Figures 8 and 9
show the comparison between the pre-defined displacement field, simulation results and experimental
measurements, as function of the longitudinal co-ordinate, and respectively at the points y=12 mm and
y=36 mm. Figures 10 and 11 show the comparison between the pre-defined displacement field, simula-
tion results and experimental measurements at the points x=60 mm and x=135 mm, as function of the
transversal y co-ordinate.
In a general way, experimental data follow much better simulation results then pre-defined ones. This
means that, on one side, experimental measurements validate the simulations models but, on the other
hand, the geometry may be too rigid, the layout of the piezoelectric actuators is not the best or the number
of piezoelectric patches are not enough to induce the structure to deform in the correct and pre-defined
way.
Fittings to the experimental and simulation data, obtained along y at the points y=12 mm and y=36 mm,
were performed in order to understand how well the measured shape of the plate is closer to the pre-
defined one. One can observe that, in general, the fitting to the simulation and experimental data follows,
approximately, the same parabolic shape as the pre-defined displacement field. Namely, for y=12 mm
2 =0.9985 for the simulation data and 2 =0.9967 for experimental points.

6
  
6LPXODWLRQ 6LPXODWLRQ
3UHB'HILQHG 3UHB'HILQHG
([SHULPHQWDO ([SHULPHQWDO

7UDQVYHUVH'LVSODFHPHQW>PP@

7UDQVYHUVH'LVSODFHPHQW>PP@
 

 

 

 
           
/RQJLWXGLQDO3RVLWLRQ>[/@ /RQJLWXGLQDO3RVLWLRQ>[/@

Figure 8. Comparison between the pre-defined Figure 9. Comparison between the pre-defined
displacement field, simulation results and ex- displacement field, simulation results and ex-
perimental measurements for y=12 mm. perimental measurements for y=36 mm.


6LPXODWLRQ
3UHB'HILQHG
 ([SHULPHQWDO

7UDQVYHUVH'LVSODFHPHQW>PP@

7UDQVYHUVH'LVSODFHPHQW>PP@

 

 

 

6LPXODWLRQ
  3UHB'HILQHG
([SHULPHQWDO


      
     
/RQJLWXGLQDO3RVLWLRQ>\/@
/RQJLWXGLQDO3RVLWLRQ>\/@

Figure 10. Comparison between the pre- Figure 11. Comparison between the pre-
defined shape, simulation results and experi- defined shape, simulation results and experi-
mental measurements for x=60 mm. mental measurements for x=135 mm.

Simulation results: optimal layout of piezoelectric actuators

The procedure developed so far assumed a specific layout of actuators and determined the optimal actua-
tion voltages with respect to that layout. However the optimisation of the actuator location for achieving
a desired shape can be advantageous in order to minimise the error in the deflection. The extension of the
GA optimisation code in order to include the best actuator location with respect to the fitness function
was done. The optimisation problem consists now in finding a set of actuation voltages and the position,
within the finite element mesh of the structure, of a limited number of piezoelectric actuators, subjected
to certain values of saturation voltages, such that the desired shape of the structure is obtained. The
optimisation algorithm was modified in order to accommodate two different types of information: the
location of each piezoelectric element and the voltage needed to apply to each of them. So, the first part
of each chromosome consisted in a binary variable with the information of a presence (1) or absence
(0) of an actuator. This sequence was as long as the number of elements of the finite element mesh of
the structure. The second part was a sequence of real numbers representing the actuation voltages. A
population size of 20 individuals, a crossover probability of 0.75, a mutation probability of 0.15 and a
maximum number of generations of 10,000 was considered. The choice of an initial population was
made based on educated guesses.

7
Clamped-free aluminium plate
This simulation aims at finding the location and the voltages of a fixed number of piezoelectric patches,
glued in the clamped-free aluminium plate, in order to deform it according to the functions shown in
Figures 12 and 13.

-4 -4
x 1 0 x 1 0

2 .5 5

2 4

1 .5 3
-z [m ]

-z [m ]
1 2

0 .5 1

0 0
0 .0 5 0 .0 5
0 .0 4 0 .0 4
0 .0 3 0 .0 3
0 .0 2 0 .1 0 .0 2 0 .1
0 .0 1 0 .0 5 0 .0 1 0 .0 5
y [m ] 0 0 y [m ] 0 0
x [m ] x [m ]

Figure 12. Case1: Pre-defined displacement Figure 13. Case2: Pre-defined displacement
field (x) = 0:0012x + 0:003x2 . field (x; y ) = 0:0015x + 0:01x2 + 0:002xy .

An aluminium plate, 144 mm long, 48 mm wide and 0.5 mm thick, together with a maximum number
of piezoelectric elements (PX5-N from Philips Components) limited to 4 with 24 mm long, 12 mm wide
and 0.3 mm thick was considered. The saturation voltages were limited to -100 Volt and +200 Volt.
The finite element model of the plate consisted in 24 elements (6 longitudinal and 4 transversal). Each
chromosome had a length of 28 variables; 24 binary values corresponding to a piezoelectric (1) or non
piezoelectric (0) element and 4 real values corresponding to the actuation voltages. Figure 14 show the
results of the optimisation procedure concerning the location of the piezoelectric elements for case 1 and
case 2 and Table 5 the corresponding optimal actuation voltages.

Figure 14. Optimal location of the 4 piezoelectric actuators for case1 and case2.

Table 5. Results concerning the optimal placement of the piezoelectric actuators, within the aluminium
plate, and the corresponding actuation voltages obtained by GA.

Piezoelectric Elements Vpiezo1 [V ] Vpiezo2 [V] Vpiezo3 [V] Vpiezo4 [V]

Case 1 1 ; 7 ; 13 ; 19 57.5 45.9 43.4 52.9
Case 2 1 ; 3 ; 19 ; 23 88.2 25.7 120.5 82.0

Since there was no penalty, in the fitness function, of the increase of the weight due to the piezoelectric

8
actuators the final optimal result, for both cases, assumes that all the four piezoelectric actuators are
contributing to induce deformations in the structure. Case 1 consists in a bending movement of the
whole plate independent of the y co-ordinate. The final result assumes that all piezoelectric actuators
are located near the clamped edge of the plate where the strain energy and the effect of the actuators is
maximum. In case 2 the plate is intended to deform in bending and in torsion. This torsion movement can
only be achieved, with this kind of piezoelectric material, if the piezoelectric actuators are displaced in a
non symmetric manner in the plate or tilted relative to the longitudinal axis. The optimal result assumes
that 2 piezoelectric actuators are located near the clamped edge, where their influence is maximum, and
located opposite to each other, with different applied voltages, while the other two are also located in
opposite sides but this time not symmetrically to each other and again with different actuation voltages,
producing all together this bending plus torsion effect.
The comparison, along x and y directions, between the pre-defined shape and the shape obtained with
the piezoelectric actuator layout of Figure 14 and with the actuation voltages of Table 5 are shown in
Figures 15 and 16 for case 1 and in Figures 17 and 18 for case 2.
 
7UDQVYHUVH'LVSODFHPHQW>PP@

7UDQVYHUVH'LVSODFHPHQW>PP@
 

 

 

 
3UHB'HILQHG 3UHB'HILQHG
6LPXODWLRQ 6LPXODWLRQ
 
           
/RQJLWXGLQDO3RVLWLRQ>[/@ /RQJLWXGLQDO3RVLWLRQ>\/@

Figure 15. Case1: Pre-defined displacement Figure 16. Case1: Pre-defined displacement
field and simulation results (y=24 mm). field and simulation results (x=144 mm).

 
7UDQVYHUVH'LVSODFHPHQW>PP@
7UDQVYHUVH'LVSODFHPHQW>PP@

 

 

 

 
3UHB'HILQHG 3UHB'HILQHG
6LPXODWLRQ 6LPXODWLRQ
 
           
/RQJLWXGLQDO3RVLWLRQ>[/@ /RQJLWXGLQDO3RVLWLRQ>\/@

Figure 17. Case2: Pre-defined displacement Figure 18. Case2: Pre-defined displacement
field and simulation results (y=24 mm). field and simulation results (x=144 mm).

In both cases a very good agreement was found between the pre-defined shape and the simulation results.
As expected, case 1 corresponds to pure bending movement with no torsion. For case 2 (Figure 18) the
torsion effect is visible and a difference less than 1% is found between the slope of the simulation results

9
(=0.0118) and the pre-defined one (=0.0117). Contrary to the case concerning the experimental results
of an aluminium plate with 4 piezoelectric actuators glued on it in an non optimal manner, there are no
significant discrepancies between pre-defined and simulation results. This means that this new layout of
piezoelectric elements is ”more” optimal than the previous one.

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 experimentally
verified. Throughout this paper only glued piezoelectric elements are considered, however the developed
methodology also supports embedded ones. Shape control was applied to a beam and plate structure
with different boundary conditions. A good agreement was found between simulation and experimental
results. The square root of the error between the nodal pre-defined displacement and the achieved dis-
placement was considered as the objective function. With the aim of decreasing the errors between the
pre-defined shape and the obtained shape of a structure the optimisation algorithm was modified in order
to consider not only the determination of optimal voltages but also to optimise the best location of the
piezoelectric actuators. The obtained results validate the adopted methodology.

REFERENCES
1. E. Crawley and J. Luis, “Use of piezoelectric actuators as elements of intelligent structures,” AIAA
Journal 25(10), pp. 1373–1385, 1987.
2. D. Koconis, L. Kollar, and G. Springer, “Shape control of composite plates and shells with embedded
actuators i. voltages specified,” Journal of Composite Materials 28(5), pp. 415–458, 1994.
3. D. Koconis, L. Kollar, and G. Springer, “Shape control of composite plates and shells with embedded
actuators ii. desired shape specified,” Journal of Composite Materials 28(3), pp. 262–285, 1994.
4. K. Chandrashekhara and S. Varadarajan, “Adaptive shape control of composite beams with piezo-
electric actuators,” Journal of Intelligent Materials and Structures 8, pp. 112–124, February 1997.
5. S. D. M. Silva, R. Ribeiro, and C. Hauviller, “High precision and stable structures for particle detec-
tors,” in SPIE’s 6th Annual International Symposium on Smart Structures and Materials, vol. 3668 -
Part Two, pp. 1017–1025, Newport Beach, CA, March 1999.
6. R. Ribeiro, S. D. M. Silva, J. Rodrigues, and M. Vaz, “Genetic algorithms for optimal design and
control of adaptive structures,” in SPIE’s 7th Annual International Symposium on Smart Structures
and Materials, to be published, Newport Beach, CA, March 2000.
7. D. E. Goldberg, Genetic Algorithms in Search, Optimization, and Machine Learning, Addison-
Wesley Publishing Company, Inc., 1989.
8. Z. Michalewicz, Genetic Algorithms + Data Structures = Evolution Programs, Springer, 1996.
9. M. Gen and R. Cheng, Genetic algorithms and engineering design, John Wiley & Sons Inc.

10