Professional Documents
Culture Documents
discussions, stats, and author profiles for this publication at: https://www.researchgate.net/publication/242072723
CITATIONS READS
40 171
5 authors, including:
Shuang Cong
University of Science and Technology of China
206 PUBLICATIONS 1,037 CITATIONS
SEE PROFILE
Some of the authors of this publication are also working on these related projects:
All content following this page was uploaded by Shuang Cong on 15 March 2014.
Abstract: In this paper, the dynamic controller design problem of a redundant planar 2-dof
parallel manipulator is studied. Using the Euler-Lagrange equation, we formulate the dynamic
model of the parallel manipulator in the joint space and propose an augmented PD controller with
forward dynamic compensation for the parallel manipulator. By formulating the controller in the
joint space, we eliminate the complex computation of the Jacobian matrix of joint angles with
end-effector coordinate. So with less computation, our controller is easier to implement, and a
shorter sampling period can be achieved, which makes the controller more suitable for high-
speed motion control. Furthermore, with the combination of static friction model and viscous
friction model, the active joint friction of the parallel manipulator is studied and compensated in
the controller. Based on the dynamic parameters of the parallel manipulator evaluated by direct
measurement and identification, motion control experiments are implemented. With the
experiments, the validity of the dynamic model is proved and the performance of the controller is
evaluated. Experiment results show that, with forward dynamic compensation, the augmented PD
controller can improve the tracking performance of the parallel manipulator over the simple PD
controller.
rapidly are not available. But the performance of space, and the complex computation of the Jacobian
kinematic controllers is limited, for the neglect of matrix of the joint angles with end-effector coordinate
joints coupling and the nonlinear dynamics of parallel is eliminated. With less computation, the controller
manipulators. The synchronous control method [10] proposed here is easier to implement, and a shorter
can be used to solve the problem of joints coupling sampling period can be achieved. This makes the
and improve the accuracy of the trajectory tracking, controller more suitable for high-speed motion control.
but the nonlinear dynamics is still an unsolved Furthermore, with the combination of static friction
problem for kinematic controllers, especially under model and viscous friction model, the active joint
high-speed motion control. friction of the parallel manipulator is compensated in
Unlike the kinematic control strategy, the dynamics the controller, while only the viscous friction of joints
of parallel manipulators is taken into account in the was considered in the previous controllers designed
dynamic control strategy. So the nonlinear dynamics by Chen [16] and Yiu [4]. With trajectory tracking
of parallel manipulators can be compensated, and the experiments, the validity of the dynamic model is
higher tracking performance can be achieved with proved and the performance of the controller is
dynamic controllers. Based on approximated linear evaluated. Experiment results show that, with forward
dynamic models, dynamic controllers were proposed dynamic compensation, the augmented PD controller
for parallel manipulators with improved tracking can improve the tracking performance of the parallel
performance [11-13]. Although the controller is easy manipulator over simple PD controller.
to implement by adopting the approximated linear The paper is organized as follows. In Section 2, the
model of parallel manipulators, the effect of the structure of the manipulator is described and the
controller is limited in a small region of the kinematics is formulated. In Section 3, with the Euler-
configuration space for the inaccurate compensation Lagrange equation, we formulate the dynamics of the
of nonlinear dynamics. To solve this problem, full parallel manipulator in the joint space. In Section 4,
nonlinear dynamic models were adopted in the we formulate the active joint friction of the parallel
controller design in [14-16] and adaptive methods manipulator with the combination of static friction
were also used to improve the tracking performance of and viscous friction. By least square method, we
parallel manipulators [4,17,18]. Furthermore, to gain a identify the unknown friction parameters and other
deep insight of the parallel manipulator dynamics, Liu unknown dynamic parameters of the parallel
developed an unified geometric approach for the manipulator. In Section 5, based on the dynamic
modeling and the control of parallel manipulators [19], model in Section 3 and the friction model in Section 4,
and Aghili proposed a more generalized method to we propose a simple PD controller and an augmented
solve the dynamic problem of parallel manipulators PD controller with forward friction compensation for
[20]. All these works have placed solid foundation for the parallel manipulator. With the two controllers,
the further research of dynamic control strategy. linear and circular trajectory tracking experiments are
The control problem of parallel manipulators with performed and experiment results are reported too.
redundant actuators is also of great interest for Section 6 concludes the paper with several important
researchers. For redundant actuated parallel remarks.
manipulators, infinite feasible solutions of control
torques can be obtained to track a single desired 2. MECHANISMS AND INVERSE
trajectory. Among the feasible solutions, the best one KINEMATICS
may be picked out to minimize a predefined optimal
function, so additional optimal control objective can The parallel manipulator to be controlled is Googol
be achieved during the trajectory tracking. In literature, Tech. Ltd’s GPM2002, which is a redundant planar 2-
redundant actuation is used to minimize the control dof parallel manipulator. As shown in Fig. 1, the
torque [15,16], achieve optimal load distribution GPM2002 consists of 6 links denoted as
among actuators [21], improve the manipulability of Lai , Lbi , i = 1, 2,3, 3 active joints located at A1 , A2 , A3
parallel manipulators [22], obtain desired end-effector
stiffness [23] and avoid the transmission backlash [24]. and 3 passive joints locate at B1 , B2 , B3 respectively.
In this paper, the dynamic controller design The end-effector of the parallel manipulator coincides
problem of a redundant planar 2-dof parallel with O in Fig. 1.
manipulator is studied. Using the Euler-Lagrange According to Fig. 1, a reference frame is estab-
equation, we formulate the dynamic model of the lished in the workspace of the parallel manipulator
parallel manipulator in the joint space and propose an and the unit of the frame is millimeter. With the
augmented PD controller for the parallel manipulator. reference frame, the zero positions of joint angles are
Compared with the controllers proposed by Kock [21] all defined as the positive direction of the X axis of the
and Chen [16] which were formulated in the frame, and the positive directions of angles are all the
workspace, our controller is formulated in the joint anticlockwise direction. The active joints of the
Modeling, Identification and Control of a Redundant Planar 2-DOF Parallel Manipulator 561
⎛ di ⎞
θ ai = arctg 2 ( x − xai , y − yai ) + arccos ⎜ ⎟,
⎝ 2l ⎠ (1)
i = 1, 2,3.
θb ⎧α = J a + ma ra2 + mb l 2
La ⎪⎪
B
⎨ β = J b + mb rb
2
A (9)
θa ⎪γ = m lr .
x ⎪⎩ b b
M θ + Cθ = τ − f .
⎧ 1 2 1
(
⎪⎪ KEa = 2 J aθ a + 2 ma xca + yca
2 2
) (11)
⎡ α1 γ1cab1 0 ⎤ dH (θ ) ∂H (θ )
0 0 0 = θ = A (θ )θ = 0. (19)
⎢ 0 α2 0 0 γ 2 cab 2 0 ⎥⎥ dt ∂θ
⎢
⎢ 0 0 α3 0 0 γ 3cab3 ⎥ Then matrix A can be formulated as follows:
M =⎢ ⎥,
⎢γ 1cab1 0 0 β1 0 0 ⎥
⎢ 0 γ 2cab 2 0 0 β2 0 ⎥ ⎡ −lsa1 lsa 2 0 −lsb1 lsb 2 0 ⎤
⎢ ⎥ ⎢ lc −lca 2 0 lcb1 −lcb 2 0 ⎥⎥
⎢⎣ 0 0 γ 3cab3 0 0 β3 ⎥⎦ A = ⎢ a1 .
⎢ −lsa1 0 lsa3 −lsb1 0 lsb3 ⎥
(16) ⎢ ⎥
⎣ lca1 0 −lca 3 lcb1 0 −lcb3 ⎦
⎡ 0 0 0
⎢ (20)
⎢ 0 0 0 With (15), (16), (17), and (20), the dynamics of the
⎢ 0 0 0 GPM2002 is established in the joint space.
C=⎢
⎢ −γ 1sab1θa1 0 0
⎢ 4. IDENTIFICATION OF DYNAMIC
0 −γ 2 sab 2θa 2 0
⎢ PARAMETERS
⎢⎣ 0 0 −γ 3 sab3θa 3
γ s θ
1 ab1 b1 0 0⎤ For the GPM2002, the dynamic parameters of links
⎥ such as masses mai, mbi, moments of inertia relative to
0 γ 2 sab 2θb 2 0⎥ the mass centers Jai, Jbi, distances between mass
0 0 γ 3 sab3θb3 ⎥ centers and joints rai, rbi and length l, can be measured
⎥. directly. But the constraint forces and the joint
0 0 0 ⎥
0 0 0 ⎥ frictions in (15) are still unknown. In this section, the
⎥ unknown constraint force AT λ is eliminated by
0 0 0 ⎥⎦ projection method, and the active joint friction is
(17) modeled with the combination of static friction model
In (16) and (17), symbols cabi , i = 1, 2,3 are defined and viscous friction model. By formulating the
as cabi = cos (θ ai − θbi ) , i = 1, 2,3, and symbols sabi , i dynamics of the GPM2002 as a group of linear
equations about dynamic parameters, the unknown
= 1, 2,3 are defined as sabi = sin (θ ai − θ bi ) , i = 1, 2,3 . friction parameters and other unknown dynamic
As shown in (15), the dynamics of the GPM2002 parameters of the GPM2002 are identified with least
is formulated in the joint space and redundant squared method.
variables are involved in the formulation, so the
constraint force AT λ is included explicitly to 4.1. Elimination of constraint forces
guarantee the closed-loop constraints of the parallel The constraint force magnitude λ in (15) is
manipulator are satisfied at every instant. Here matrix unknown, and it is difficult to be measured directly.
A is the differential of the closed-loop constrained Fortunately this unknown term can be eliminated by
equations, and multiplier λ denotes the magnitude projection method [19].
of constraint forces. With (15), the magnitude of constraint force can be
Adopting the symbols defined in Section 2, one calculated through following equations:
( )
can formulate the closed-loop constraints of the −1
GPM2002 as following equation: λ = AM −1 AT ( )
AM −1 M θ + Cθ − (τ − f ) . (21)
564 Yao-Xin Zhang, Shuang Cong, Wei-Wei Shang, Ze-Xiang Li, and Shi-Long Jiang
Substituting (21) into (15), one can obtain following In (25), symbols Wi , ti , p are defined as follows:
equations:
p = [k fv1 fv 2 f v3 d1 + f s1 d1 − f s1
( )
⎛ −1
⎞
⎝
−1 T
⎜ I − A AM A
T
⎠
(
AM −1 ⎟ M θ + Cθ − (τ − f ) = 0. ) d2 + f s 2 d2 − fs2 d3 + f s 3 d3 − f s 3 ] ,
T
(26)
(22) ⎡ 6
With (22), one can eliminate constraint forces by Wi = ⎢ ∑ Pij D j Pi1θa1 Pi 2θa 2 Pi 3θa3
projecting the dynamic model onto the image space of ⎢⎣ j =1 (27)
( )
−1
matrix P = I − AT AM −1 AT AM −1. The parame- ui1 li1 ui 2 li 2 ui 3 li 3 ] , i = 1," , 6,
3
ters of P consist of the dimension parameters of the ti = ∑ Pijτ aj , i = 1," , 6. (28)
parallel manipulator and the inertia parameters of j =1
links, all of which can be measured directly. So the
matrix P can be calculated directly and constraint For simplicity, parameter combinations di + f si
forces can be eliminated with (22).
and di − f si are viewed as unknown parameters to
4.2. Model of joint frictions be identified as shown in (26). In (27), the coefficients
In (22), joint friction vector f is still an unknown uik , lik , k = 1, 2,3 are determined by following rules:
term. Here, static friction model and viscous friction u = P , l = 0 when θ > 0 and u = 0, l = P
ik ik ik ak ik ik ik
model are adopted to formulate the joint frictions.
when θak < 0. In (28), only the input torque of
Furthermore, the zero drift of the control board that
results a nonzero output torque for motors is active joints is considered, for the input torque of
considered too. For the GPM2002, the passive joints passive joints always equals zero.
are well lubricated, so the frictions of passive joints There are 10 parameters to be identified in (25) and
can be ignored and only the frictions of active joints 2 independent equations can be got for each sampling
are considered. The frictions of active joints can be point. So a group of linear equation about the
formulated as following equations: unknown parameters can be got with the sampling
data of a continuous trajectory. By solving the linear
( )
fi = di + sign θai f si + f viθai , i = 1, 2,3. (23) equations with least squared method, the unknown
parameters can be identified.
Here, symbols di , i = 1, 2,3 refer to the zero drifts of
4.4. Parameter identification experiment
control board, symbols f si , i = 1, 2,3 refer to the
The dynamic parameters of links can be measured
static frictions, and symbols f vi , i = 1, 2,3 refer to directly, as shown in Table 1.
viscous friction coefficients. To identify the friction parameters and the unknown
dynamic parameters of the GPM2002, the end-
4.3. Formulation of parameter identification effector of the parallel manipulator is driven to track a
In (22) and (23), the dynamic parameters of links circular trajectory. Based on (25), a group of linear
can be measured directly and only the friction equations about the unknown parameters can be
parameters have to be identified. Besides, the unit of obtained with the sampling data of trajectory, and the
output torque of the motors has to be transformed to
the unit of numerical control value, so one more Table 1. Dynamic parameters of the links.
unknown unit transformation coefficient has to be Distance
identified. Based on (22), following equation can be Moment of
between
got: Mass Length inertia relative
the mass
(Kg) (m) to the mass
( )
P M θ + Cθ + f − τ = P ⋅ D ⋅ k + P ⋅ f − P ⋅ τ = 0. center and
the joint (m)
center (Kg*mP2P)
(24)
La1 1.2525 0.2440 0.1156 0.0124
In (24), symbol k refers to the unknown unit
transformation coefficient, while symbol D refers to La 2 1.3663 0.2440 0.0657 0.0122
the sum of the inertia term and the coriolis term of the
La3 1.3663 0.2440 0.0657 0.0122
manipulator. Let P = ⎡⎣ Pij ⎤⎦ and D = ⎡⎣ D j ⎤⎦ ,
6×6 6×1 Lb1 1.0771 0.2440 0.1621 0.0098
one can formulate (24) as a group of linear equations
about the unknown dynamic parameters. Lb 2 0.4132 0.2440 0.1096 0.0036
unknown parameters can be identified with least controller, and ei refers to the tracking error of
squared method. In the identification experiment, the active joints which can be formulates as follows:
velocity of joints is got by the numerical
differentiation of joint positions, and the acceleration ei = θˆai − θ ai , i = 1, 2,3. (30)
of joints is got by numerical differentiation of joint
velocity. Since the acceleration of joints is calculated 5.2. Augmented PD controller
by two-time numerical differentiations, the joint The structure of augmented PD controller is shown
accelerations are quite noisy. A low-pass filter is in Fig. 4.
adopted to filter the velocity signal before the For the GPM2002, the dynamics including the
calculation of acceleration. friction at active joints dominates the dynamics of the
Based on the parameter values in Table 1, the parallel manipulator, so we ignore the dynamics
unknown parameters of the GPM2002 are identified including the friction at passive joints. Then
with MATLAB program and results are shown in computation of the command torque of motors can be
Table 2. simplified further, which can be calculated by
following equations:
5. CONTROLLER DESIGN AND MOTION
CONTROL EXPERIMENTS τ i = K p ei + K d ei + kDi + fi , i = 1, 2,3, (31)
Here symbols di, i = 1,2,3 refer to the zero drifts of the adopted to evaluate the performance of the simple PD
motion control board, fsi, i = 1,2,3 refer to the static controller and the augmented PD controller [16].
frictions of active joints, fvi, i =1,2,3 refer to the
∑ ( ( xˆi − xi ) + ( yˆi − yi ) )
viscous friction coefficient of active joints. N
1 2 2
RSME = (34)
N i =1
5.3. Trajectory tracking experiment results
With the controllers proposed above, motion In (34), symbols xˆi , yˆi refer to the coordinates of the
control experiments are implemented. In the
experiments, the end-effector of the GPM2002 is ith interpolation point of desired trajectory and xi , yi
driven to track a linear trajectory and a circular refer to the coordinates of corresponding point of real
trajectory both for 2 times, once with lower velocity trajectory.
and once with higher velocity. The control system The starting point of linear trajectory is (150,300)
runs on a Pentium III CPU at 333MHz with the and the ending point is (300,150). The velocity profile
sampling period 2ms. In the controller, the unit of of desired linear trajectory is a T-curve. First the end-
feedback signal is rad for joint position and rad/2ms effector of the GPM2002 is driven to track the line
for joint velocity. The value of the proportion gain Kp with maximum velocity 0.2m/s and acceleration 4m/s2P P,
of simple PD controller is 800000, and derivative gain then with maximum velocity 0.5m/s and acceleration
Kd is 5000000. For the augmented PD controller, the 10m/s2P P. Tracking errors of the end-effector are shown
same proportion gain and derivative gain with simple in Figs. 5 and 6, while the RSMEs of the linear
PD controller are adopted. trajectory tracking experiments are shown in Table 3.
Furthermore, root-square mean error (RSME) is
1 2
PD Controller PD Controller
Augmented PD Controller Augmented PD Controller
0.5 1
error [mm]
error [mm]
0 0
-0.5 -1
0 200 400 600 800 1000 0 100 200 300 400
time [ms] time [ms]
0 0
error [mm]
error [mm]
-0.5 -1
-1 -2
0 200 400 600 800 1000 0 100 200 300 400
time [ms] time [ms]
error [mm]
End-effector
0.7140 0.1111 -2
velocity 0.2m/s
End-effector -3
1.1011 0.2560
velocity 0.5m/s
-4 PD Controller
Augmented PD Controller
The starting point of circular trajectory is -5
0 500 1000 1500 2000 2500
(216.5,180) and center is (216.5,250). The end- time [ms]
effector of the GPM2002 is driven to track the circle
(a) Tracking error in X direction.
for two times, first with constant velocity 0.2m/s and
then with constant velocity 0.5m/s. Tracking errors of 2
the end-effector are shown in Figs. 7 and 8, while the PD Controller
Augmented PD Controller
RSMEs of the circular trajectory tracking experiments
are shown in Table 4. 1
1
error [mm]
0 -1
error [mm]
-2
-1 0 500 1000 1500 2000 2500
time [ms]
Yao-Xin Zhang received the B.S. Shi-Long Jiang received the B.S.
degree in Automation from University degree in Automation from the
of Science & Technology of China in Taiyuan Institute of Mechanical,
2002. From 2002 he has been studying Taiyuan, PRC, the M.Ph. degree in
in University of Science & Technology Information Engineering from Central
of China for the Doctor's degree. His South University of Technology,
current research interests include Changsha, PRC, and the Ph.D. degree
parallel manipulator and multi-axes in Electrical and Electronic Engineer-
synchronous control. ing from the Hong Kong University of
Science and Technology, in 1991, 1994, and 2000,
respectively. He is a Senior Engineer in Googol Technology
(Shenzhen) Limited. His research interests include robotics,
tactile sensor, multifingered manipulation, parallel robot
manipulation, and motion control.