Beruflich Dokumente
Kultur Dokumente
Mechatronics
journal homepage: www.elsevier.com/locate/mechatronics
a r t i c l e
i n f o
Article history:
Received 28 April 2011
Accepted 3 February 2012
Available online 8 March 2012
Keywords:
Piezoelectric actuator
Flexure-based mechanism
Real-coded genetic algorithm
Clonal selection algorithm
Particle swarm optimization
System identication
a b s t r a c t
Compensations for cross-axis coupling effect and hysteretic nonlinearity of a novel XY piezo-actuated
positioning stage are presented in this study. The piezo-actuated stage utilizes a monolithic exure-based
mechanism (FBM) to achieve translations in X- and Y-axes instead of using stacked mechanisms. A hysteresis model with crossover term is proposed to alleviate the cross coupling effect between X- and Ystages during precision positioning tasks. System identications using real-coded genetic algorithm
(RGA) and clonal selection algorithm (CSA) are compared with particle swarm optimization (PSO). The
results show that PSO provides better performance than the others. Therefore, a feedforward controller
with cross-axis coupling compensation is studied and the used for the piezo-actuated FBM to enhance
the precision of the coarse positioning stage. The experimental results conrm that the proposed controller can achieve precision tracking tasks with submicron precision.
2012 Elsevier Ltd. All rights reserved.
1. Introduction
The demand for ultra-precision positioning systems has increased in the high-precision, high-speed and long-stoke machining industries, especially in semiconductor and ultra-precision
machining during recent years. A ball-screw based mechanism is
the most popular in industrial applications, because it has a wide
diversity of application for different precision and load requirements. However, achieving sub-micrometer accuracy over a longstroke is difcult for ball-screw based mechanisms, because backlash and friction between the nut and the ball-screw. To solve the
above problem, combining a coarse positioning stage with a micropositioning mechanism is a practical method, where the coarse
stage with a ball-screw covers long range traveling with an
accuracy of a few micrometers and the micro-positioning mechanism takes care of small residual range of sub-micrometer accuracy. Sakuta et al. developed a dual servo mechanism where the
PA was mounted on the friction drive slide to achieve an accuracy
positioning task by using a kind of inchworm movement [1].
Okazaki et al. developed a dual-servo mechanical stage by using
a ne and coarse servo positioning system, where a laser interferometer and gap gage were applied to measure the displacement of
the stage [2]. The servo system was controlled with task sharing in
the frequency and displacement range. Lee and Kim developed a
dual servo stage, where a 3-DOF linear motor and 3-DOF PZT actuated stage were used to drive a motion stage by a PID scheme [3].
Corresponding author. Tel.: +886 2 2771 2171; fax: +886 2 2711 1401.
E-mail addresses: cjlin@ntut.edu.tw (C.-J. Lin), t7618006@ntut.edu.tw (P.-T. Lin).
0957-4158/$ - see front matter 2012 Elsevier Ltd. All rights reserved.
doi:10.1016/j.mechatronics.2012.02.001
615
616
[45,46], the advantage is that no cross-coupling effects appear between the X- and Y-stages because the two PAs are located in the
different PMs. The disadvantages of using the stacked XY PMs are
larger coplanar errors for biaxial positioning and its mechanisms
size is larger than the monolithic mechanism.
Fig. 1a and b shows that the mechanical architecture of the
piezo-actuated exure-based mechanism (PAFBM) is equipped
with an optical 0.1 lm resolution encoder (Mercury 2000, made
by MicroE systems, USA) on each moving stage to provide position
feedback. Fig. 1a and b shows top and bottom views of the PAFBM.
The presented PAFBM is actuated by a PA for each axis using an
optical encoder with linear glass scales for positioning feedback.
The monolithic PAFBM is fabricated by a wire electric discharge
machine (WEDM) from a piece of A7075 material. Fig. 1 shows that
the monolithic PAFBM enables independent motion of the translation platform in the X and Y directions. The X- or Y-moving stage is
moved as an electrical voltage applied to the PA, respectively. Each
PA is a 20 lm-stoke PZTs (model PSt 150/5x5/20, manufactured by
Piezomechanik GmbH, Germany), which is actuated by a PZT driver
(voltage 20 to +150 V; model CA45, manufactured by CEDRAT,
France). Tables 1 and 2 describe the specications for the PA and
the PZT driver, respectively. The feedback signal of the optical encoder is 0.1 lm/count and is decoded by the DSP controller (model:
DS1104 PPC control board, made by dSPACE, Germany) to enable
data acquisition and real-time control. The dSPACE DS1104 has a
MPC8240 processor with PowerPC 603e core and a 250 MHz CPU
clock. The dSPACE DS1104 board is equipped with 32 MB application memory for storing the real-time programs to be executed.
Table 1
The specication of the PEA.
Type number
PSt 150/5x5/20
5 mm 5 mm
18 mm
20 lm
1800 nF
50 kHz
60 N/lm
1600 N
2000 N
Table 2
The searching range for the parameters of the standalone system.
Parameters
Range
m1 (kg)
m2 (kg)
b1, b2 (N s/m)
k1, k2 (N/m)
d1, d2 (m/V)
a1, a2
b1, b2
c1, c2
0.150.17
0.060.07
010
78 107
05 107
01
01
01
K12,b12
K11,b11
K22,
K22,b22
K21,b21
1,b21
Kp1,bp1
To obtain the standalone model, suppose the X-stage and Ystage are not actuated simultaneously. Therefore, the cross-coupling effect between the X- and Y-axis can be neglected. To simplify
the complexity of the mechanism, the structure was modeled in
terms of the lump mass discrete system, as described in Fig. 2.
The modeling of each PA actuated FBM stage can be derived as
follows.
For X-stage, the dynamic model can be formulated as follows.
m2
K23,b23
23,b23
K24,b24
K24,
Kp2,bp2
K13,b13
K14,b14
m1
Fig. 2. The modeling of a nested XY PZT FBM in terms of the lump mass discrete
system.
10mm
m1 x1 b1 x_ 1 k1 x1 k1 d1 U 1 h1 ;
where the parameter b1 = bFBM1 + bPzt1 and k1 = kFBM1 + kPzt1 are the
equivalent viscous damping coefcient and stiffness factor of the
whole system, respectively; d1 = d33 nc, d33 is the piezoelectric con-
10mm
Fig. 1. (a) The top-view of the PAFBM stage and (b) the bottom-view of the PAFBM stage.
617
Fig. 3. (a) The deformation of the PAFBM and (b) the displacement vector plot with 30 magnication due to the cross-coupling effect.
stant of each ceramic layer (lm/V), nc is the number of ceramic layers, U1 is the applied voltage. Due to the hysteresis phenomena of
PA, the relation between the actuated force and the applied voltage
is nonlinear and the nonlinear hysteretic term is denoted as h1.
fPzt1 = k1(d1 U1 h1) is the actual force with respect to the applied
voltage. If the dynamic of hysteresis can be estimated by a nonlinear hysteresis observer, then this system can be linearized via a
feedforward controller as follows.
U FF1 t
^1 t
xd t h
;
d1
m1 x1 b1 x_ 1 k1 x1 k1 xd D;
^_ 1 a1 d1 U_ 1 b jU_ 1 jh
^1 c U_ 1 jh
^1 j;
h
1
1
x1
m11
0
x_ 1
b
k
0
11
11
0
m22 x2
0 b22 x_ 2
0
k11 k12 d1 U 1 h1
;
k21 k22 d2 U 2 h2
0
k22
x1
x2
618
K 11
Ebd
L
X axis
10
8
7
6
5
4
3
2
20
40
60
80
Voltage (V)
100
6
5
4
3
Standalone
Cross-coupling
Cross-coupling
Standalone
Y axis
10
Displacement ( m)
Displacement ( m)
11
120
10
20
30
40
50
60
70
80
90
Voltage (V)
Fig. 5. (a) The hysteresis loop of the X-axis PZT stages in the standalone mode and the experiential result with cross-coupling effects and (b) the hysteresis loop of the Y-axis
PZT stages in the standalone mode and the experiential result with cross-coupling effects.
619
s
XN xh i xi2
Min
Jx
;
i1
m1 ;b1 ;k1 ;d1 ;a1 ;b1 ;c1
N
Subject to
Equality constraints:
m1 x b1 x_ k1 x k1 d1 U 1 h1 ;
9a
9b
Inequality constraints:
0 6 m1 6 m1
0 6 b1 6 b1
k1 6 k1 6 k1
0 6 d1 6 d1
9c
0 6 a1 6 a1
0 6 b1 6 b1
0 6 c1 6 c1 ;
where J(x) is the objective function; the xh(i) represents the experimental data of the hysteric loop via the measurement at the ith
sampling time in Fig. 4a; x(i) are the calculated data via Eqs. (9a)
and (9b) at the ith sampling time; N is the total sampling number
from the experimental results. In this optimal problem, the objective function J(x) represents the root mean squared error (RMSE) between the experimental results and the simulation data. If the value
of J(x) can be much smaller, then the estimated hysteresis model
will be more accurate.
3.2. Identication based on RGA
Genetic algorithms (GAs) are procedures used to obtain solutions for optimal search problems through application of the principles of evolutionary biology. GAs were rst proposed by John
Holland in the 1960s [62] and have become widely used in various
disciplines. GA is a stochastic technique based on the mechanism
of natural selection and natural genetics. A chromosome is a string
of symbols. On one hand, it is usually a binary bit string, but not
necessarily. On the other hand, the GA representing the genes directly as real numbers is called real-code genetic algorithm
(RGA). The chromosomes evolve through successive iterations,
called generations. During each generation, the chromosomes are
evaluated, using some measures of tness. To create the next generation, new chromosomes, called offspring, are formed by either
merging two chromosomes from current generation using crossover operator or modifying a chromosome using a mutation operator. A new generation is formed by selecting, according to the
tness values, some of the parents and offspring and rejecting others so as to keep the population size constant. Fitter chromosomes
have higher probabilities of being selected. After several generations, the algorithm converges to the best chromosome, which
hopefully represents the optimum or suboptimal solution to the
problem. In this study, a real-coded genetic algorithm (RGA) is
adopted to identify system parameters. The feature of real-coded
is easy to formulate the optimization problems without coding
and decoding processes. RGA is able to exploit local continuities
of the function and uses different mutation and crossover techniques to implement real representations. The identication based
on RGA was implemented as follows.
(1) Stochastic uniform parent selection: It lays out a line in
which each parent corresponds to a section of the line of
length proportional to its scaled value. The algorithm moves
along the line in steps of equal size. At each step, the algorithm allocates a parent form the section it lands on. The rst
step is a uniform random number less than the step size.
(2) Reproduction with elitist strategy and crossover fraction:
Reserve two elites that are guaranteed to survive to the next
generation. The crossover method used is a combination of
two vectors v1, v2 as follows.
v 01 g v 1 1 g v 2 ;
v 02 1 g v 1 g v 2 ;
where g is a uniformly distributed random variable between 0
and 1.
(3) Gaussian mutation: The Gaussian mutation adds a random
number taken from a Gaussian distribution with mean 0 to
each element of an individuals vector to create a new offspring. In the simplest case, the three parameters are
mutated with the same Gaussian distribution as follows.
v new
v i Dv ;
i
10
T
where vi is a vector [mi, bi, ki, di, ai, bi, ci] and Dv is distributed
according to the following probability distribution function (PDF):
1
1 Dv 2
;
pDv p exp
2 r2
2pr
11
xi xi1 1 shrink
i
;
generations
12
Min FX;
X2Q
13
where F: Q ? R, F(X) = J(x), X = [mi, bi, ki, di, ai, bi, ci], Q = {X e Rn:
li 6 xi 6 ui, i = 1, . . . , n}, Q is the feasible domain as the box constraints in Eq. (9c), x satises Eqs. (9a) and (9b), and n = 7 in this
study. All chromosomes carry out crossover and mutation to compose a population and they are evaluated according to the tness
function. The individuals with better tness values (smaller RMSE)
will survive and the individuals with bad tness values will be given up. The best chromosome is usually allowed to propagate to
the next generation, such as an approach called an elitist strategy.
The left individuals with better RMSE compose a new population
and return to the second step. RGA searches the best solution over
generations in these steps until the converging index is satised.
3.3. Identication based on PSO
In mid 1990s Eberhart and Kennedy enunciated an alternative
solution to the complex non-linear optimization problem by emulating the collective behavior of bird ocks, particles, the Boids
method of Craig Reynolds and socio-cognition and called their
14
xid t 1 xid t v id t 1:
15
100
CSA
PSO
RGA
RMSE
brainchild the particle swarm optimization (PSO) [63]. PSO is a relatively recent heuristic search method based on the idea of collaborative behavior and swarming in biological populations. It is a
multi-agent parallel search technique. PSO is similar to the GA in
population operations, but PSO has no evolution operators such
as crossover and mutation. In PSO, the potential solutions, or called
particles, y through the problem space by following the current
optimum particles. The detailed information will be given in following sections. Particles are conceptual entities that y through
the multi-dimensional search space. At any particular instant, each
particle has a position and a velocity. The position vector of a particle with respect to the origin of the search space represents a trial
solution of the search problem. First, a population of particles is
*
initialized with random positions marked by vectors x i and ran*
dom velocities v i . The population of such particles is called a
swarm. Each particle has two state variables, its current position
*
*
x i t and its current velocity v i t. Each particle is also equipped
*
with a small memory comprising its previous best position p t,
which is the personal best experience and the best of all particles
so far, and the best value so far in the group among the best posi*
tion g t, which is referred to as the globally best particle in the entire swarm. The global version of PSO is implemented as follows:
10-1
10
10
10
Generation
0
10
CSA
PSO
RGA
RMSE
620
-1
10
10
10
10
Generation
Fig. 6. Comparisons of convergence characteristics in RCGA, PSO and CSA methods
for BoucWen models for (a) the X-axis stage and (b) the Y-axis stage.
neighborhood best position are near each other, the particle performs a local search. The constriction coefcient method therefore
balances the need for local and global search depending on the so*
*
cial conditions in place. Initially, the settings for p t and g t are
*
*
*
p 0 g 0 x 0 for all particles. Once all particles are initialized, an iterative process optimizes the positions and velocities of
all particles according to the following recursive equations:
V id t 1 vv id c1 u1 pid t xid t c2 u2
g id t xid t;
xid t 1 xid t v id t 1;
16
17
where
2
q with / c1 c2 :
j4 / /2 4/j
18
Since coefcient c1 contributes to the self-exploration of a particle, coefcient c1 is considered the self-condence of the particle.
The contribution of coefcient c2 to motion of the particles depends on the motion of all the particles in the preceding program
iterations, so its denition as swarm condence is apparent.
Coefcients /1 and /2 represent uniformly distributed random
numbers in the interval [1]. The rst iteration of the algorithm is
completed when the velocities and position for the next time step
t + 1 are calculated.
621
RGA
PSO
CSA
0.1612
0.1566
0.0655
0.0000
0.0000
0.0000
7.5211E7
7.0028E7
7.0001E7
8.1813E8
8.1947E8
8.1929E8
0.6388
0.5590
0.5628
0.0125
0.0148
0.0156
0.0693
0.0552
0.0552
Table 4
The identied values of BoucWen hysteresis model for the Y-axis FBM.
Parameters
RGA
PSO
CSA
0.0673
0.0631
0.0651
0.0000
0.0000
0.0000
7.0723E7
7.6963E7
7.2732E7
1.1065E7
1.1064E7
1.1062E7
0.5231
0.5243
0.5313
0.1108
0.1111
0.1089
0.0000
0.0000
0.0039
Table 5
The RMSE and computing time based on X-axis stage of FBM.
Method
RMSE (lm)
RGA
PSO
CSA
0.092309
0.091625
0.092331
3969
1993
30384
Table 6
The RMSE and computing time based on Y-axis stage of FBM.
Parameters
RMSE (lm)
RGA
PSO
CSA
0.147957
0.147950
0.147972
3793
2179
30630
X axis
11
10
8
7
6
5
4
3
7
6
5
4
3
2
Experiment
20
40
60
80
Voltage (V)
100
Experiment
1
0
Y axis
10
Displacement (m)
Displacement ( m)
(1) Initialization: The rst step of the CSA, which involves preparing an antibody pool of xed size. This pool is separated
into two components: a memory antibody section and a
remaining antibody pool. The memory antibody section
becomes representative of the algorithms solution eventually and remaining antibody pool is used for introducing
additional diversity into the immune system.
(2) Loop: The algorithm executes over iterations to make the
immune system expose to all known antigens. A single
round of exposure or iteration is referred to as a
generation.
120
10 20 30 40 50 60 70 80 90
Voltage (V)
Fig. 7. (a) The hysteresis loops of the X-axis PEA stages obtained by experimental data and the response for the identied BoucWen model using PSO method and (b) the
hysteresis loops of the Y-axis PEA stages obtained by experimental data and the response for the identied BoucWen model using PSO method in the standalone mode.
622
Table 7
The searching range for the parameters of the system for the XY PZT stage.
Parameters
Searching range
a1
a2
01
01
01
01
01
01
Obtained in
Obtained in
Obtained in
Obtained in
Obtained in
Obtained in
Obtained in
9000000
09E6
Obtained in
0.519979
0.863758
0.012794
0.295187
0.052803
0
8.19E08
1.11E07
0.1566
0.0631
0.000000
0.000000
7.00E + 07
3.23E + 06
4.15E + 06
7.70E + 07
b1
b2
c1
c2
d1
d2
m11
m22
b11
b22
k11
k12
k21
k22
Case
Case
Case
Case
Case
Case
Case
1
1
1
1
1
1
1
Case 1
action between X- and Y-axes is discussed as the two PAs are actuated synchronously. From Fig. 5, it is obvious that there exists
cross-coupling effects for the XY PZT system when the triangle
waveforms with the amplitude of 130 and 90 (V) are synchronously applied to the X-axiss and Y-axiss PAs. Therefore, Eq. (6)
should be used to describe the dynamics of XY PZT stage instead
of Eq. (2). For the individual X-axis or Y-axis stage, the mass,
damping coefcients, spring constants and piezoelectric constants
(m11, m22, b11, b22, k11, k22, d1, d2) can be obtained in Case I and the
cross-coupling effects do not inuence them. Therefore, the optimization problem can be simplied to search eight parameters
for the cross-coupling XY PZT system. On one hand, the force generated by the X-axis or Y-axis PAs affect both two FBM stage
through the stiffness k12 and k21. On the other hand, the Bouc
Wen model with two outputs with two inputs are determined by
six parameters a1, a2, b1, b2, c1, and c2. Therefore, the optimization
for XY PZT system is formulated as follows with LMI constraints as
follows.
s
RMSEx 2 RMSEy 2
Min
Jx
:
k12 ;k21 ;a1 ;b1 ;c1 ;a2 ;b2 ;c2
2
19
Subject to
Equality constraints:
x1
m11
0
x_ 1
b11 0
k
11
0
m22 x2
0 b22 x_ 2
0
k
k12 d1 U 1 h1
;
11
k21 k22 d2 U 2 h2
"
h_ 1
h_ 2
"
0
k22
x1
x2
20a
20b
Inequality constraints:
0 6 k12 6 k12
0 6 k21 6 k21
0 6 a1 6 a1
0 6 b1 6 b1
0 6 c1 6 c1
20c
0 6 a2 6 a2
0 6 b2 6 b2
0 6 c2 6 c2 ;
where J(x) is the objective function; the RMSEx and RMSEy represent
the root mean square errors with respect to experimental data of
623
Fig. 8. The block-diagram of the XY PZT FBM system with considering the cross-coupling effect.
the hysteric loop via the measurement for the X- and Y-axes, respectively. Table 7 describes the searching range for the inequality constraints and the optimal system parameters obtained by the PSO
method. Fig. 8 shows the block-diagram of the proposed model with
cross-axis coupling term. Fig. 9a and b compares the simulation results using the standalone model with the experimental results
with cross-coupling effects; there exist large modeling errors for
the standalone model due to cross-coupling effects between the
X- and Y-axes PAs.
To verify the effectiveness of the proposed cross-coupling model, the optimal parameters obtained by PSO are used to Eq. (6).
Fig. 9a and b shows that the proposed cross-coupling model has
less tracking errors than the standalone model with respect to
the experimental results. Table 8 shows that the modeling errors
for the standalone and the cross-coupling models; the modeling
error for the proposed cross-coupling model has less errors than
the standalone model for each axis. Therefore, the simulation results verify the effectiveness of the proposed model.
U 1 t
xd t
:
d1
21
624
X axis
10
8
7
6
5
4
3
1
0
7
6
5
4
3
Experiment
Cross-coupling model
Standalone model
20
40
60
Y axis
10
Displacement ( m)
Displacement (m)
11
80
100
Experiment
Cross-coupling model
Standalone model
120
10
20
30 40
50
60 70
80
90
Voltage (V)
Voltage (V)
Fig. 9. The comparison between the experimental result, the response of the standalone model, and the cross-coupling model for (a) the X-axis PZT FBM and (b) the Y-axis
PZT FBM.
Table 8
The root-mean-square modeling errors for the standalone and the cross-coupling
models.
RMSE (lm)
X axis
Y axis
Standalone model
Cross-coupling model
0.154339
0.087902
0.379272
0.118803
where e1(t) = xd(t) x(t) represents the positioning error of the system with the feedforward compensating of the inverse hysteresis.
Therefore, Fig. 9 describes the block diagram of the PA-actuated
FBM system. Substituting Eq. (24) into Eq. (2) gives
b1
k1
k1
x
d1 U 1 h1
x_
m1
m1
m1
Z
b1
k1
k1 d1
k1
x
K P1 e1 t K I1 e1 tdt
xd
x_
m1
m1
m1
m1
k1 ^
h1 h1 :
m1
x
When the system has no control and the hysteresis is taken into
consideration, the actual displacement of the PEA, xPEA(t), can be
described as follows due to the hysteresis effect.
xPEA t d1 U 1 t h1 t;
22
U FF1 t
^ t
xd t h
1
;
d1
23
^
where ht
is the nominal hysteresis obtained by the BoucWen
model, whose parameters are identied by the PSO in Table 7.
The ideal control performance is achieved without feedback based
on IMC. However, there still exists modeling errors due to the
non-symmetry of hysteresis or uncertainties and external disturbances, so that the feedback controller is necessary to enhance
the robustness of the system and improve the tracking performance; the feedforward controller is combined to the PI feedback
controller as follows,
U 1 t U FF1 t U FB1 t;
24
where
U FB1 t K P1 e1 t K I1
Feedforward
control
x d ( t ) + e (t )
1
~ 1 t h1 t
Dene the modeling uncertainty of the hysteresis, h
R
^
h1 t, and new state variables z1(t) = e1(t)dt, z2(t) = e1(t), and
z3 t e_ 1 t. Then, the closed-loop error dynamics of the X-axis
stage can be described as follows.
3
2 3
z1
z_ 1
b1
k1 ~
6_ 7
6 7
h1 ;
x_ d
4 z2 5 A 4 z2 5 B xd
m1
m1
z3
z_ 3
2
6
A4
2 3
0
6 7
1 7
5 and B 4 0 5:
mb11
1
0
k1 dm1 1K I1
k1 km1 d11 K P1
z2
X_ A X n;
K P1 + K I 1
1
s
U 1 (t )
27
k1
m1
28
h1 ( t )
+
26
where
V X T PX;
e1 tdt;
25
d1 U 1 (t ) h1 (t )
f Pzt1 (t )
1
ms 2 + bs + k
Fig. 10. The block diagram of the proposed control for the PZT XY FBM system.
x (t )
625
Fig. 11. The embedded Simulink code using the dSPACE controller for the real-time control architecture.
X Axis
Open loop
Feedforward
Feedforward+feedback
Displacement (mm)
Error (m)
0.5
-0.5
-1
-1.5
0.5
1.5
2.5
Time (second)
Fig. 12. The tracking errors of the three controllers for X-axis stage.
Fig. 14. The response of the micro-stepping stage with respect to the sinusoidal
reference with 1 mm amplitude.
Open loop
Feedforward
Feedforward + feedback
0.6
0.4
Error (m)
Time (sec)
Y Axis
0.8
0.2
where P e R33 is a symmetric positive-denite matrix. By construction the Lyapunov function is nonzero except at the origin. Differentiating Eq. (28) with respect to time gives:
0
-0.2
-0.4
-0.6
-0.8
-1
0.5
1.5
2.5
Time (second)
Fig. 13. The tracking errors of the three controllers for Y-axis stage.
29
626
PA AT P Q :
30
1
l kXk;
2
where dene
satises
31
Q
l kkmin
and the derivative of Lyapunov function
max P
V_ X T PAX X T AT PX 2X T Pn
Error (m)
knk 6
X T QX 2X T Pn
6 X T QX 2kPXk knk
32
0.4
0.3
0.2
0.1
0
-0.1
-0.2
-0.3
Time (sec)
xd 105 t us t 2us t 1 us t 2;
Fig. 15. The tracking error of the micro-stepping stage and the positioning
reference to the PAFBM.
Error (m)
33
where us(t) is a unit step function. The parameters for the feedforward controller is obtained in Case II, the experimental results show
that the proposed feedforward controller can almost compensate
the hysteresis nonlinearities. The sample time for real-time implementation of the dSPACE controller is set as 0.001 s. Figs. 12 and
13 show the tracking errors of the three controllers for the X- and
Y-stages; the maximal tracking errors of the open-loop system are
about 1.2 to +0.7 lm for the X-stage and 0.9 to +0.8 lm for the
Y-stage.
Figs. 12 and 13 show that the proposed feedforward controller
reduces the X-axiss tracking errors from 1.2 to +0.7 lm to 0.3
to +0.2 lm and it reduces the Y-axiss tracking errors from 0.9
to +0.8 lm to 0.2 to +0.2 lm. The proposed feedforward control
based on PSO is valid according to the resulting response. The
feed-forward controller with PI feedback controller is applied to
the positioning task. The parameters for the PI feedback controller
Fig. 16. The tracking error for the integrated system for the micro-stepping stage
with compensation of the PAFBM.
are Kp = 0.8 and KI = 480 for each X-axis and Y-axis of PAFBM. Figs.
12 and 13 show that the feedforward with feedback controller has
tracking errors about 0.1 to +0.1 lm for both the X- and Y-stages.
Therefore, this case study validates the proposed method and the
positioning precision has been within 0.1 lm.
5. Conclusions
This paper presented a novel monolithic FBM actuated by the
embedded piezoelectric actuators (PAs) to achieve the translations
in the X- and Y-axes. In Case 1, the modeling identication for the
standalone PAFBM using the real-coded genetic algorithm (RGA),
clonal selection algorithm (CSA) and particle swarm optimization
(PSO) were studied; the simulation results are found that the
PSO method takes less time than RGA and CSA to obtain the optimal solution. In Case 2, the cross-coupling effect between the X and
Y axes was discussed; the performance of the standalone model
was used to make comparison with the cross-coupling model.
The results show that the proposed cross-coupling model has less
modeling error than the standalone model. In Case 3, to compensate the nonlinearity of hysteresis, the PI feedback controller with
the feed-forward control based on inverse hysteresis model was
illustrated. The real-time experimental results validated the proposed method and the positioning precision has been within
0.1 lm. In Case 4, the PAFMB with linear encoder feedback was
studied to enhance the precision of coarse positioning stages.
Experimental results validated the effectiveness of the proposed
approach.
Acknowledgments
This work is supported by the National Science Council under
Grant NSC100-2221-E-027-031 and the Ministry of Economic
Affairs, ROC under Grant 98-EC-17-A-16-S1-127.
References
[1] Sakuta S, Okawa K, Ueda K. Experimental studies on ultra-precision positioning
an inchworm movement method using ne and coarse positioning. Int J Jpn
Soc Prec Eng 1993;27:23540.
[2] Okazaki Y, Asano S, Goto T. Dual-servo mechanical stage for continuous
positioning. Int J Jpn Soc Prec Eng 1993;27:1723.
[3] Lee CW, Kim SW. An ultra precision stage for alignment of wafers in advanced
microlithography. Prec Eng 1997;21:11322.
[4] Chen JS, Dwang IC. A ballscrew drive mechanism with piezo-electric nut for
preload and motion control. Int J Mach Tools Manuf 2000;40:51326.
[5] Pahk HJ, Lee DS, Park JH. Ultra precision positioning system for servo motorpiezo actuator using the dual servo loop and digital lter implementation. Int J
Mach Tools Manuf 2001;41:5161.
[6] Abramovitch DY, Andersson SB, Pao LY, Schitter G. A tutorial on the
mechanisms, dynamics, and 28 control of atomic force microscopes. In:
American control conference; 2007. p. 3488502.
[7] Clayton GM, Tien S, Leang KK, Zou Q, Devasia S. A review of feedforward
control approaches in nanopositioning for high-speed SPM. J Dyn Syst Measur
Control 2009;131:10119.
[8] Devasia S, Eleftheriou E, Reza Moheimani SO. A survey of control issues in
nanopositioning. IEEE Trans Control Syst Technol 2007;15:80223.
[9] Ryu JW, Gweon DG, Moon KS. Optimal design of a exure hinge based XYHz
wafer stage. Prec Eng 1997;21:1828.
[10] Chang SH, Chung KT, Hon CC. An ultra-precision XYHz piezo-micropositioner
part I: design and analysis. IEEE Trans Ultrason Ferroelect Freq Control
1999;46:897905.
[11] Chang SH, Chung KT, Hon CC. An ultra-precision XYHz piezo-micropositioner
part II: experiment and performance. IEEE Trans Ultrason Ferroelect Freq
Control 1999;46:90612.
[12] Uchino K. Piezoelectric actuator and ultrasonic motors. Norwell (MA): Kluwer
Academic Publishers; 1997.
[13] Catalog PI. Theory and application of piezo actuators and PZT nanopositioning
systems. Waldbronn (Germany): Physik Instrument; 1998.
627
[14] Lee CW, Kim SW. An ultraprecision stage for alignment of wafers in advance
microlithography. Prec Eng 1997;21:11322.
[15] Leang KK, Fleming AJ. High-speed serial-kinematic AFM scanner: design and
drive considerations. Asian J Control 2009;11:14453.
[16] Schitter G, Thurner PJ, Hansma PK. Design and input-shaping control of a novel
scanner for high-speed atomic force microscopy. Mechatronics 2008;18:
2828.
[17] Comstock R. Charge control of piezoelectric actuators to reduce hysteresis
effects. United States patent #4.263,527. Assignee: The Charles Stark Draper
Laboratory, Cambridge, MA; 1981.
[18] Goldfarb M, Celanovic N. Modeling piezoelectric stack actuators for control of
micromanipulation. IEEE Trans Control Syst 1997;17:6979.
[19] Tsai JC, Chen CL, Lee YH, Yang HY, Hsu MS, Chen KH. Modied hysteretic
current control for improving transient response of boost converter. IEEE Trans
Circ Syst 2011;58:196779.
[20] Mayergoyz ID. The Preisach model for hysteresis. Berlin: Springer; 1991.
[21] KrasnoselskI MA, Pokrovski AV. Systems with hysteresis. New
York: Springer-Verlag; 1980.
[22] Mayergoyz ID. Dynamic Preisach models of hysteresis. IEEE Trans Magn
1988;24:29257.
[23] Stepanenko Y, Su CY. Intelligent control of piezoelectric actuators. In: Proc IEEE
conf deci control; 1998. p. 42349.
[24] Bouc R. Forced vibration of mechanical systems with hysteresis. In: 4th Int
conf on nonlinear oscillation; 1967.
[25] Wen YK. Method of random vibration of hysteretic systems. J Eng Mech
1976;102:24963.
[26] Zhang X, Huang Y, Liu J, Wang X, Gao F. A method identifying the parameters
of BoucWen hysteretic nonlinear model based on genetic algorithm. Intell
Process Syst 1997:6025.
[27] Xu QS, Li YM. Dahl model-based hysteresis compensation and precise
positioning control of an XY parallel micromanipulator with piezoelectric
actuation. J Dyn Syst Measur Control 2010;132:041011104101112.
[28] Chua LO, Bass SC. A generalized hysteresis model. IEEE Trans Circ Theory
1972;19:3648.
[29] Leite JV, Sadowski N, Kuo-Peng P, Batistela NJ, Bastos JPA. The inverse Jiles
Atherton
model
parameters
identication.
IEEE
Trans
Magn
2003;39:1397400.
[30] Croft D, Devasia S. Vibration compensation for high speed scanning tunneling
microscopy. Rev Sci Instrum 1999;70:46005.
[31] Croft D, Shedd G, Devasia S. Creep hysteresis and vibration compensation for
piezoactuators: atomic force microscopy application. ASME J Dyn Syst Meas
Control 2001;123:3543.
[32] Schitter G, Thurner PJ, Hansma PK. Design and input-shaping control of a novel
scanner for high-speed atomic force microscopy. Mechatronics
2008;18:2828.
[33] Perez H, Zou Q, Devasia S. Design and control of optimal scan-trajectories:
scanning tunneling microscope example. ASME J Dyn Syst Meas Control
2004;126:18797.
[34] Fleming AJ, Wills A. Optimal input signals for band-limited scanning systems.
In: Proceedings of the 17th IFAC world congress; 2008: p. 1180510.
[35] Barrett RC, Quate CF. Optical scan-correction system applied to atomic force
microscopy. Rev Sci Instrum 1991;62:13939.
[36] Francis BA, Wonham WM. The internal model principle for linear
multivariable regulators. Appl Math Optim 1975;2:17094.
[37] Francis BA, Wonham WM. The internal model principle of control theory.
Automatica 1976;12:45765.
[38] Pao, LY, Butterworth JA, Abramovitch DY. Combined feedforward/feedback
control of atomic force microscopes. In: Proceedings of the 2007 American
control conference; 2007: p. 350915.
[39] Li Y, Bechhoefer J. Feedforward control of a closed-loop piezoelectric
translation stage for atomic force microscope. Rev Sci Instrum 2007;78:
013702.
[40] Leang KK, Devasia S. Feedback-linearized inverse feedforward for creep,
hysteresis, and vibration compensation in AFM piezoactuators. IEEE Trans
Control Syst Technol 2007;15:92735.
[41] Aphale SS, Devasia S, Moheimani SOR. High-bandwidth control of a
piezoelectric nanopositioning stage in the presence of plant uncertainties.
Nanotechnology 2008;19:125503.
[42] Salapaka S, Sebastian A, Cleveland JP, Salapaka MV. High bandwidth nanopositioner: a robust control approach. Rev Sci Instrum 2002;73:323241.
[43] Ha JL, Kung YS, Fung RF, Hsien SC. A comparison of tness functions for the
identication of a piezoelectric hysteretic actuator based on the real-coded
genetic algorithm. Sens Actuat 2006;132:64350.
[44] Jang MJ, Chen CL, Lee JR. Modeling and control of a piezoelectric actuator
driven system with asymmetric hysteresis. J Franklin Inst Eng Appl Math
2009;346:1732.
[45] Lin CJ, Chen SY. Evolutionary algorithm based feedforward control for
contouring of a biaxial piezo-actuated stage. Mechatronics 2009;19:
82939.
[46] Lin CJ, Yang SR. Precise positioning of piezo-actuated stages using hysteresisobserver based control. Mechatronics 2006;16:41726.
[47] Products P-280 and P-762, Physik Instrumente product catalog,
micropostioning, nanaopositioning, nanoautomation: solutions for cuttingedge technologies; 2001.
[48] Fischer FL. Symmetrical 3 DOF compliance structure. US patent #4447048;
1981.
628
[49] Smith AR, Gwo S, Shih CK. A new high resolution two-dimensional
micropositioning device for scanning probe microscopy. Rev Sci Instrum
1994;64:32169.
[50] Dagalakis NG, Kramer JA, Amatucci E, Bunch R. Kinematic modelling and
analysis of planer micro-positioner. In: Proceedings of ASPE 2001 annual
meeting; 2001: p. 1358.
[51] Agilent Technologies NanoStepper. MIT technology review, J7220 data sheet.
[52] Bednorz JG. Piezoelectric XY positioner. US patent #4520570; 1985.
[53] M-850, F-206 HEXAPOD. Physik Instrumente product catalog; 2001.
[54] Culpepper ML. Multiple degree of freedom complaint mechanism. US patent
application #20030086751; 2003.
[55] Dagalakis NG, Amatucci EG. Six degree of freedom micropositioner. US patent
#6484602, 2002.
[56] Davies PA. Positioning mechanism. US patent #6193226; 2001.
[57] Jokiel B, Benavides GL, Bieg LF, Allen JA. Planer and spatial three degree of
freedom micro stages in silicon MEMS. In: Proceedings of ASPE 2001 annual
meeting; 2001: p. 325.
[58] Eom TB, Kim JY. Long range stage for the metrological atomic force
microscope. In: Proceedings of ASPE 2001 annual meeting; 2001. p. 1569.
[59] Hitachi Ltd. Ultra-precision two-dimensional moving apparatus. US patent
#4575942; 1986.
[60] IBM Corp. Two-dimensional positioning apparatus, US patent 5059090;
1991.