Professional Documents
Culture Documents
a
School of Electro-Mechanical Engineering, Xidian University, Xi’an 710071, China
b
Beijing Astronomical Observatory, Chinese Academy of Sciences, Beijing 100012, China
Received 1 April 2001; accepted 28 June 2001
Abstract
A new singularity analysis method for general six degree-of-freedom (DOF) Stewart plat-
form using genetic algorithm (GA) is proposed in this paper. The Jacobian matrix of Stewart
platform is first deduced, then the square of determinant of the Jacobian matrix is selected as
the objective function, and the minimal of this objective function is searched in the workspace
of Stewart platform by the GA. The singularity of Stewart platform depends on this minimal
objective function: if this value is zero, the singularity of Stewart platform will take place,
otherwise, the Stewart platform is singularity-free. The effectiveness of this new genetic sin-
gularity analysis method is validated by the singularity analysis of a six-DOF fine-tuning
Stewart platform for the next generation large radio telescope. The results have shown that the
fine-tuning Stewart platform is singularity-free, which has laid a solid base for the requirement
of high precision trajectory tracking for the next generation large radio telescope.
Ó 2002 Elsevier Science Ltd. All rights reserved.
Keywords: Stewart platform; Mechatronics; Singularity analysis; Genetic algorithm; Large radio telescope
1. Introduction
Radio astronomers from around the world have focused recently on the need for
a new radio telescope which is 2 orders of magnitude greater than the biggest op-
erational telescopes, increasing sensitivity by a factor of 100, which would enable a
direct observation of the formation and evolution of galaxies from gases in the
*
Corresponding author. Fax: +86-29-823-2281.
E-mail addresses: yxsu@mail.xidian.edu.cn (Y.X. Su), byduan@xidian.edu.cn (B.Y. Duan).
0957-4158/03/$ - see front matter Ó 2002 Elsevier Science Ltd. All rights reserved.
PII: S 0 9 5 7 - 4 1 5 8 ( 0 1 ) 0 0 0 5 1 - 4
414 Y.X. Su et al. / Mechatronics 13 (2003) 413–425
universe [1]. To fulfill this goal, the next generation large radio telescope (LT, now it
is SKA, the square kilometer array) project was undertaken throughout the world
[2,3]. To make full use of the karst formation in Guizhou Province of China, a large
spherical radio telescope array of Arecibo telescope [4], with about 30 large spherical
reflector antennas and each antenna with the diameter of up to 300–500 m, was
proposed in China in 1996 [2].
For the sake of overcoming the shortcomings of the original design of Arecibo
telescope, Duan [5] proposed a completely new design project integrating mecha-
tronics and optics technologies. In this project, the fixed back-up structure weight
of 800 ton of Arecibo telescope as thrown away and replaced by a moving cabin
structure which weighed only 20–30 ton, and was driven by six large span cables to
implement trajectory tracking of the feed. It is well known that, for the large span
cable-controlled system, it is difficult to guarantee the high precision required by
trajectory tracking (state error less than 4 mm). Based on this design project, Su and
Duan [6] developed a fine-tuning Stewart platform to fulfill the high requirements of
feed tracking for a large spherical radio telescope, and the large span cables only
manage the coarse trajectory tracking control. The feasibility of this modified
mechatronics design project has been validated with numerical simulation, just as
shown in Fig. 1.
In recent years, great interest has been devoted to Stewart platform, even though
the first industrial applications were restricted only to flight simulators [7]. Both an
increase in applications and the advantages of parallel geometry over serial open
chains have been strong incentives to the development of the theory and practice of
Stewart platform [6–17]. Among these, the singularity of Stewart platform is still an
open problem [11–16]. The physical notion of a singularity in kinematics refers to
configurations in which a mechanism’s number of degrees-of-freedom (DOF)
changes instantaneously. For serial chains that have an end-effector frame attached
to the final link, a kinematic singularity is a configuration in which the end-effector
Fig. 1. Modified optomechatronics design of the next generation large radio telescope.
Y.X. Su et al. / Mechatronics 13 (2003) 413–425 415
loses one or more DOF of motion. This means that there are certain directions in
Cartesian space along which the device is instantaneously not free to move, no
matter what the joint rates are. In contrast, in parallel manipulator, these singular
configurations result in the gain of one or more DOF. This means that the structure
is no longer rigid and there are certain directions in Cartesian space along which the
device cannot support a force or movement [11]. Gosselin and Angeles [12] classified
the singularity of the Stewart platform into the following three categories:
1. the set of points where different branches of the inverse-kinematic problem meet,
2. the set of points where different branches of the forward-kinematic problem meet,
and
3. the case when the Jacobian matrix is indeterminate.
2. Jacobian matrix
The general six-DOF fine-tuning Stewart platform for the next generation large
radio telescope is illustrated in Fig. 2. It consists of a moving platform, a base
platform, and six linear actuators, each of which is composed of a ballscrew
416 Y.X. Su et al. / Mechatronics 13 (2003) 413–425
Fig. 2. Fine-tuning Stewart platform for the next generation large radio telescope.
assembly mounted axially with a dc motor, connecting the moving platform and base
platform together with spherical and universal joints, respectively.
Motion of the moving platform with respect to the base platform is produced by
driving the motors to shorten or extend the actuator lengths, and the proper coor-
dination of the actuator length trajectories enables the moving platform to perform
the complex trajectory with high accuracy.
The following frame is developed. The inertial reference frame O-XYZ is fixed on
the base, and the movable coordinate system p-xyz is assigned to the moving plat-
form, with their origins at the centers of the two circular platforms, respectively, as
shown in Fig. 2.
ðxp ; yp ; zp Þ and ðw; h; /Þ are defined as the position of the original p of the movable
frame p-xyz referred to the inertial reference system O-XYZ and the angular rotation
to the Z; Y ; X axes of the inertial reference frame O-XYZ, separately.
For the ith electrical linear actuator li , the coordinate of the center Ai which is
connected to the fixed base platform in the fixed frame O-XYZ is Ai ðXi ; Yi ; Zi Þ, and
the coordinate of the center Bi which is connected to the moving platform in the
movable coordinate system p-xyz is Bi ðxi ; yi ; zi Þ.
With the rotation coordinate transformation, the relative coordinate Ci ðXri ;
Yri ; Zri Þ of the center Bi ðxi ; yi ; zi Þ with respect to the fixed frame O-XYZ can be ex-
pressed as
T T T
ðXri ; Yri ; Zri Þ ¼ R ðxi ; yi ; zi Þ þ ðxp ; yp ; zp Þ ; ð1Þ
where R is the following rotation matrix:
2 3
cos w cos h sin w cos h sin h
R ¼ 4 sin w cos / þ cos w sin h sin / cos w cos / sin w sin h sin / cos h sin / 5
sin w sin / cos w sin h cos / cos w sin / þ sin w sin h cos / cos h cos /
ð2Þ
Therefore, given the position of the moving platform referred to the base, the ap-
propriate length of the electrical linear actuator can be calculated by the following
expression:
Y.X. Su et al. / Mechatronics 13 (2003) 413–425 417
1=2
2 2 2
li ¼ ðXi Xri Þ þ ðYi Yri Þ þ ðZi Zri Þ ð3Þ
The Jacobian matrix of Stewart platform defines the relationship between the
velocity of six legs in joint space and the velocity of moving platform in workspace
[6,11]. The following notation is specified: P_ is the velocity of the moving platform,
which can be defined as the velocity of the center of the moving platform as
T
V ¼ ½_xp ; y_ p ; z_ p (mm/s), and the angular velocity of the moving platform to the three
Z; Y ; X axes as x ¼ ½w_ ; h_; /_ T (rad/s), so P_ can be described by
P_ ¼ ½V ; x ¼ ½_xp ; y_ p ; z_ p ; w_ ; h_; /_
T T
ð4Þ
and l_ is the velocity of linear actuator in joint space, which can be expressed as
T
l_ ¼ ½l_ 1 ; l_ 2 ; l_ 3 ; l_ 4 ; l_ 5 ; l_ 6 : ð5Þ
P_ ¼ J l_ ; ð6Þ
For simplicity, the following symbols are used to develop the Jacobian matrix:
Ai1 ¼ Xi Xri ;
Bi1 ¼ Yi Yri ; ð8Þ
Ci1 ¼ Zi Zri ;
Ai2 ¼ swchxi cwchyi ;
ð9Þ
Ai3 ¼ cwshxi þ swshyi þ chzi ;
Bi2 ¼ ðcwch swshs/Þxi ðswc/ þ cwshs/Þyi ;
Bi3 ¼ cwchs/xi swchs/yi þ shs/zi ; ð10Þ
Bi4 ¼ ðsws/ þ cwshc/Þxi ðcws/ þ swshc/Þyi chc/zi ;
Ci2 ¼ ðcws/ þ swshc/Þxi ðsws/ cwshc/Þyi ;
Ci3 ¼ cwchc/xi þ swchc/yi shc/zi ; ð11Þ
Ci4 ¼ ðswc/ þ cwshs/Þxi þ ðcwc/ swshs/Þyi chs/zi :
418 Y.X. Su et al. / Mechatronics 13 (2003) 413–425
3. Real-coded GAs
GAs are heuristic search algorithms based on the mechanics of natural selection
and natural genetics. GAs were initially proposed by Holland [18], and have been
widely used in a variety of fields such as combinatorial optimization problems [18–
22]. GAs are optimization methods inspired by natural evolution, and perform a
multiple directional search by maintaining a population of potential solutions. The
population-to-population approach attempts to make the search avoid local optima.
The population undergoes a simulated evolution: At each generation the relatively
good solutions are reproduced, while the relatively bad solutions are eliminated.
GAs use probabilistic transition rules to select solutions for reproduction and
elimination so as to guide their search toward regions of the search space with likely
improvement. This leads to stronger and stronger generations which evolve towards
the solution of the problem. Fig. 3 shows the searching procedure of simple GAs.
To apply GAs to solve the optimal issue, one has to consider the following issues
[18]:
How to encode a solution of the problem into a chromosome is the key issue for
GAs. Choosing an appropriate encoding strategy to represent the candidate solutions
to the problem at hand is the foundation for applying GAs to solve real world prob-
lems [18]. Strategy selection depends highly on the nature of the problem variables.
At present, there are two schemes to accomplish this definition of the chromo-
some, that is, the binary coded scheme and real-coded scheme [21]. Since, any ar-
Y.X. Su et al. / Mechatronics 13 (2003) 413–425 419
Because the real-coded scheme is used, the solution space coincides with the
chromosome space. The strategy to produce the initial population is a probabilistic
420 Y.X. Su et al. / Mechatronics 13 (2003) 413–425
methodology [18]. Considering the diversification of the initial population, the initial
values of the position and orientation should be distributed in the solution space as
evenly as possible.
Just as mentioned above, the singularity of general six-DOF Stewart platform can
be analyzed by the minimal of the determinant of Jacobian matrix. For the re-
quirement of the GAs, the minimal objective function can be determined as follows:
where S represents the workspace of the general Stewart platform, and pi stands
T
for the corresponding position and orientation, that is P ¼ ½p1 ; p2 ; p3 ; p4 ; p5 ; p6 ¼
T
½xp ; yp ; zp ; w; h; / .
GAs are optimal searching methods to find the maximum fitness of the individual
chromosome, so it is necessary to transform the minimal objective function into the
maximum fitness function [18]. The maximum fitness function can be expressed by
the following transformation:
1
F ðP Þ ¼ ; ð14Þ
f ðP Þ þ k
where k is a preset constant, which avoids the operation of GA cannot proceed when
the minimum value of f ðP Þ is 0.
A genetic operator is the key to implement the GAs, and it includes the following
three basic operators: selection, crossover and mutation [18].
This way, the solutions that are far away from each other are not compared to
each other and diversity among feasible solutions is maintained.
4. Results
In the iteration of the real-coded GA, the corresponding position and orientation
to the maximum fitness function among the generations are shown in Figs. 4 and 5,
respectively. With the development of the iteration of the GA, the minimal deter-
minant of the Jacobian matrix of the fine-tuning Stewart platform for the next
generation large radio telescope is shown in Fig. 6.
It can be seen from the above figures that the genetic singularity analysis method
for the general Stewart platform is highly efficient and can be accomplished in
424 Y.X. Su et al. / Mechatronics 13 (2003) 413–425
practice with convenience. Only 10 iterations are occurring the minimal objective
function can be converged.
In the whole workspace, the minimal determinant of the Jacobian matrix of the
fine-tuning Stewart platform is expressed as follows:
detðJ Þmin ¼ 9:30578: ð20Þ
It can be seen from the above minimal determinant of the Jacobian matrix that the
designed fine-tuning Stewart platform for the next generation large radio telescope is
singularity-free, which has laid a solid base for the high precision requirement of the
trajectory tracking.
5. Conclusions
1. A new genetic singularity analysis method for general six-DOF Stewart platform
is proposed in this paper. Once the workspace of the Stewart platform is deter-
mined, then the minimal determinant of the Jacobian matrix can be searched
by GAs. The singularity of Stewart platform depends on this minimal value: if
it is zero, then there are certain singularities, otherwise, the Stewart platform is
singularity-free.
2. The genetic singularity analysis method for general Stewart platform is highly ef-
ficient and convenient for practical applications with an example of the singular-
ity analysis of the fine-tuning Stewart platform for the next generation large radio
telescope.
3. The designed fine-tuning Stewart platform for the next generation large radio tele-
scope is demonstrated to be singularity-free using the above genetic singularity
analysis method, which has laid a solid base for the high precision requirement
of the trajectory tracking.
Acknowledgements
The authors would like to express their appreciation for the financial support of
the National Natural Science Foundation with nos. 59675040 and 50075065 and the
Key Project of Innovation Program in Chinese Academy of Sciences, China. The
authors are also very thankful to the anonymous referees for the numerous remarks
and suggestions which have led to significant improvements of this paper.
Y.X. Su et al. / Mechatronics 13 (2003) 413–425 425
References