Professional Documents
Culture Documents
1-11, 2011
Copyright 2011, TSI Press
Printed in the USA. All rights reserved
1. INTRODUCTION
Although cranes are very important systems for handling heavy goods, automatic cranes are
comparatively rare in industrial practice [4][21], because of high investment costs. The need for
faster cargo handling requires control of the crane motion so that its dynamic performance is
optimized. Specifically, the control of overhead crane systems aims to achieve both position
regulation and anti-swing control [6]. Several authors have discussed this problem, for example,
time-optimal control was considered in [3], the use of boundary conditions was developed in [2]
and [22]. Unfortunately, to increase robustness, some time optimization requirements, like zero
angular velocity at the target point [19], have to be given up. Gain scheduling has been proposed
as a practicable method [7] to increase tracking accuracy, while observer-based feedback control
was presented in [21]. Many attempts, such as planar operation [7] and assuming the absence of
friction [19], have been made to introduce simplified models for application of model-based
control [21]. Thus, a self-tuning controller with a multilayer perceptron model for an overhead
crane system was proposed in [17], while in [5] the controller consists of a combined position
servo control and a fuzzy-logic anti-swing controller.
There are two main weaknesses in applying PD control to this application: (a) The PD
controller requires suitable sensors to provide measurements of both position and velocity.
Position can be obtained very accurately by means of an encoder, while velocity is usually
measured by a tachometer, which can be expensive and is often contaminated by noise [12]; (b)
Due to the existence of friction and gravitational forces, the steady-state error is not guaranteed to
be zero [13]. It is therefore important to be able to realize PD control using only position
measurement. One possible approach is to use a velocity observer, which can be either modelbased or model-free. Model-based observers assume that the dynamics of the overhead crane are
either completely or partially known. For example, the variable structure observer in [8] needed
information about the inertia matrix to calculate the sliding mode gain. In contrast model-free
observers do not require such exact knowledge about the overhead cranes. The most popular
model-free observers are high-gain ones which can estimate the derivative of the output [20].
Recently, an observer was presented in reference [12], where the nonlinearity of the manipulator
was estimated by a static neural network. In this paper, a high-gain observer which can achieve
stability is added to regular PD control.
The anti-swing control problem involves reducing the swing of the payload while moving it
to the desired position as fast as possible [1]. One particular feedforward approach is input
shaping [23], which is an especially practical and effective method of reducing vibrations in
flexible systems. In [18] the anti-swing motion-planning problem is solved using the kinematic
model in [15]. Here, anti-swing control for a three-dimensional overhead crane is proposed, which
addresses the suppression of load swing. Nonlinear anti-swing control based on the singular
perturbation method is presented in [26]. Unfortunately, all of these anti-swing controllers are
model-based. In this paper, a PID law is used for anti-swing control which, being model-free, will
affect the position control.
Fx
xw
Fy
R
FR
xw
yw
Fx
yw
Fy
3D Crane
FR
payload
Mcg
(a)
(b)
The overhead crane system described schematically in Figure 1 (a) has the system structure
shown in Figure 1 (b). Here is the payload angle of the projection in the XY plane, is the
payload projection angle along the X-coordinate axis. The dynamics of the overhead crane are
given by [24]:
x = [x w , y w , , , R ] , (x w , y w , R )
T
1 = [Fx , Fy , 0, 0, FR ] , Fx
T
is
the
(1)
position
of
the
payload,
T
F = [ x , y , 0, 0, R ] x& , x ,
and
are frictions
factors, G ( x ) is the gravitational force, C ( x , x& ) is the Coriolis matrix and M ( x ) is the
dynamic matrix of the crane.
In (1), there are some differences from other crane models in the literature. The length of the
lift-line is not considered in [Fang], so the dimension of M is 4 4, while in [18], which also
addresses anti-swing control and position control, the dimension of M is 3 3. In [14], the
x T M& ( x ) 2C ( x, x& ) x = 0
(2)
The control problem is to move the rail in such a way that the actual position of the payload
reaches the desired one. The three control inputs Fx , Fy , FR can force the crane to the position
[xw , y w , R],
[ , ]
directly. In order to design an anti-swing control, linearization models for , are analyzed.
Because the acceleration of the crane is much smaller than the gravitational acceleration, the rope
length is kept slowly varying and the swing is not big, giving
s1 = sin , c1 = cos 1,
The approximated dynamics of
[ , ]
are then
&x&w =
Fx
Mr
, &y&w =
Fy
Mm
&& + g =
Only
angles
F
Fx
, && + g = y
Mr
Mm
(3)
Fx and Fy participate in the anti-swing control. FR does not affect the swing
, .
Fx = A1 ( x w , x& w ) + A2 ( , & )
Fy = B1 ( y w , y& w ) + B2 , &
(4)
where
model
&& + g + MA = MA
&& + g + MB = MB
1
2
r
(5)
Now if
and
B1
Mr
A2
Mr
and
B2
Mm
(5) is a second-order linear system with disturbances. Standard PID control can now be applied to
regulate and thereby producing the anti-swing controllers
t
A2 ( , & ) = k pa 2 + kda 2& + kia 2 0 ( ) d
t
B2 , & = k pb 2 + kdb 2 & + kib 2 0 ( ) d
where k pa 2 ,
(6)
and integral gains. Substituting (4) into (1), produces the position control model
]T
= A1 , B1 , 0, 0, FR
where D = A2 , B2 , 0, 0, 0 ,
controller will be designed in next section.
]T .
(7)
= K p ( x x d ) K d ( x& x& d )
where K p and K d are positive definite, symmetric and constant matrices, which correspond
to the proportional and derivative coefficients, x
d
x& d = 0.
r = ( x& x& d ) + ( x x d ) = ~
x 2 + ~
x1
x1 = ( x x
where ~
), ~
x2 = ( x& x& d ),
Because f ( x, x& , ~
x1 )
T
~
x1 = ~
x 2 , = T > 0, s = x T , x& T , ~
x1T .
is unknown, a generic fuzzy model, provided by a collection of l fuzzy
d
dt
Here
(8)
fx , f y and fz are the uncertainties (friction, gravity and coupling errors) along
the X,Y,Z -coordinate axis. i = 1, 2Ll . A total of fuzzy IF-THEN rules are used to perform the
x = [xw , y w , , , R ] 5
T
T
y (k ) = f1 , f2 , f3 = [ y1 , y 2 , y 3 ] R 3 . Here A1i ,L Ani and B1i ,L Bmi are standard
fuzzy sets. In this paper, some on-line learning algorithms are introduced for the membership
functions A ji and B ji such that the PD controller with the fuzzy compensator is stable.
By using product inference, center-average defuzzification and a singleton fuzzifier, the
output of the fuzzy logic system can be expressed as [25]
p th
y p = w pii p , p = 1, 2, 3
(9)
i =1
p
ji
B = 1.
pi
f = Wt (s )
where
the
parameter
matrix
W p = [w p1 L w pl ],
( x ) = [1 , 2 , 3 ] ,
(10)
p = 1p Ll
].
and
p T
the
data
vector
A .
p
ji
The
]T
where x = x w , y w , , , R ,
x d = xwd , y wd , 0, 0, R d , and x wd ,
(11)
fuzzy learning law. The coupling between anti-swing control and position control can be
explained as follows. For the anti-swing control (5), the position control A1 and B1 are
disturbances, which can be decreased by the integral action in PID control. Although the antiswing model (5) is an approximator, the anti-swing control (6) does not in fact use this, as it is
model-free. Hence while the anti-swing control law (6) cannot suppress the swing completely, it
can minimize any consequent vibration.
]T
For the position control (7), the anti-swing control lies in the term D = A2 , B2 , 0, 0, 0 ,
which can also be regarded as a disturbance, see Figure 2. The coupling due to anti-swing control
can be compensated by the fuzzy system. For example, in order to decrease the swing, we should
increase A2 and B2 , this means increase the disturbances of the crane, so should be
increased.
PID anti-swing
control
A2 ,B2
xw* , yw* , R*
PD position A1, B1, FR
control
xw , yw , R
Fx , Fy , FR
Crane
Fuzzy
compensator
xw , yw , R
xw , yw , R
x&1 = x 2
x& 2 = H 1 ( X , )
(12)
y = x1
where
x1 = [xw , y w , , , R ]
(13)
If the velocity vector x 2 is not measurable and the dynamics of manipulator are unknown, a
high-gain observer can be used to estimate x 2 [20]
d
dt
d
dt
where x1 , x 2
5
x1 = x 2 + 1 K1 ( x1 x1 )
x 2 = 12 K 2 ( x1 x1 )
(14)
is a
small positive parameter, and K1 and K 2 are positive definite matrices chosen such that the
K1
K 2
I
is stable. Defining the observer error as
0
~
x = x x , ~
z =~
x, ~
z = ~
x
matrix
(15)
where x = [ x1 , x 2 ] , the observer error equation can then be formed from (12) and (14)
T
T T
(16)
K1
K 2
where A =
I
,
0
d ~
z = A~
z + 2 BH1
dt
(17)
0
B = . The structure of the velocity observer is the same as in
I
[20], but a new theorem is proposed here in order to integrate the observer and the fuzzy
compensator. The PD control law (11) in combination with the state estimate from a high-gain
observer is then given by:
= K (x x d ) K (x 2 x& d ) W t ( s1 )
(18)
x 2 is of course the velocity approximation from the high-gain observer. The structure of the
anti-swing control with fuzzy compensation is shown in Figure 2.
Now
defining
~
Wt = W Wt
for
the
filtered
regulation
error
r = ( x& x& d ) + ( x x d ) , the following theorem holds. If the updating laws for the
membership functions in (fuzneu) are
d
Wt = K w ( s1 ) r1T
dt
where K w is positive definite matrix, then the PD control law with fuzzy compensation in (18)
can make the tracking error r1 stable [27].
5. EXPERIMENTAL COMPARISONS
The proposed anti-swing control for overhead crane systems has been implemented on a
InTeCo [11] overhead crane test-bed, see Figure 3. The rail is 150 cm long, and the physical
parameters for the system are as follows: the mass of rail is M r = 6.5kg , the mass of cart is
The position control law in equation (11) is discussed next. In this case there are two types of
]T
= [A1 ,L]T .
A1 is designed by (18), the anti-swing control A2 in (7) is regarded as a disturbance which will
be compensated for the fuzzy system (10). K d
should be large enough such that
K d > g1 + 1 . Since these upper bounds are not known, K d 1 = diag [80, 80, 0, 0,10] is
selected. The position feedback gain does not affect the stability, but it should be positive, and
was chosen as K p1 = diag 5, 5, 0, 0,1 .
A total of 20 fuzzy rules were used to compensate the friction, gravity and the coupling
from anti-swing control. The membership function for A ji was chosen to be the Gaussian
function
ji
(x ) = [ 1 L 20 ] .
T
(0,1) . Hence,
Wt R 520 ,
compensate the uncertainties such as friction, gravity and anti-swing coupling. Because the PID
controller has no adaptive mechanism, it does not work well for anti-swing coupling in contrast to
the fuzzy compensator which can adjust its control action. On the other hand, the PID controller is
faster than the PD control with fuzzy compensation in the case of small anti-swing coupling.
yw(m)
0.7
0.6
0.5
0.4
0.3
0.2
xw(m)
0.1
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
(rad )
0.01
0
-0.01
-0.02
-0.03
-0.04
-0.05
(rad )
-0.06
-0.15
-0.1
-0.05
0.05
0.1
0.15
0.2
Voltage(V )
0.8
0.6
Fx
0.4
Fy
0.2
0
FR
-0.2
-0.4
-0.6
-0.8
-1
time(s )
5
10
15
20
25
30
35
40
6. CONCLUSION
In this paper, the disadvantages of the popular PD control for overhead crane are overcome in
the following two ways: (I) a high-gain observer is proposed for the estimation of the velocities of
the joints; (II) a fuzzy compensator is used to compensate for gravity and friction. Real-time
10
experiments were presented comparing our stable anti-swing PD control strategy with regular
crane controllers. These showed that the PD control law with the anti-swing and fuzzy
compensations is effective for the crane system.
REFERENCES
1.
2.
3.
4.
5.
6.
7.
8.
9.
10.
11.
12.
13.
14.
15.
16.
17.
18.
19.
20.
E.M. Abdel-Rahman, A.H. Nayfeh, Z.N. Masoud, Dynamics and control of cranes: a
review, Journal of Vibration and Control, Vol.9, No.7, 863-908,2003.
J.W. Auernig and H. Troger, Time optimal control of overhead cranes with hoisting of the
payload, Automatica, Vol.23, No.4, 437-447, 1987.
J.W. Beeston, Closed-loop time optimatial control of a suspended payload-a design study,
Proc. 4th IFAC World Congress, 85-99, Warsaw Poland, 1969.
W. Blajer and K. Kotodziejczyk, Motion planning and control of gantry cranes in cluttered
work environment, IET Control Theory & Applications, Vol.1, No.5, 1370 - 1379, 2007.
S.K. Cho and H.H. Lee, A fuzzy-logic antiswing controller for three-dimensional overhead
cranes, ISA Trans., Vol.41, No.2, 235-43, 2002.
A.H.W. Chun and R.Y.M. Wong, Improving Quality of Crane-Lorry Assignments With
Constraint Programming, IEEE Transactions on Systems, Man, and Cybernetics, Part C,
Vol.37, No. 2, 268 - 277 2007.
G. Corriga, A. Giua, and G. Usai, An implicit gain-scheduling controller for cranes, IEEE
Trans. Control Systems Technology, Vol,6, No.1, 15-20, 1998.
C. Canudas de Wit and J.J.E. Slotine, Sliding observers for overhead crane manipulator,
Automatica, Vol.27, No.5, 859-864, 1991.
G. Cybenko, Approximation by superposition of sigmoidal activation function,
Math.Control, Sig Syst, Vol.2, 303-314, 1989.
Y. Fang, W.E. Dixon, D.M. Dawson, and E. Zergeroglu, Nonlinear coupling control laws
for an underactuated overhead crane system, IEEE/ASME Trans. Mechatronics, Vol.8,
No.3, 418-423, 2003.
InTeCo, 3DCrane: Installation and Commissioning Version 1.2, Krakow, Poland, 2000.
Y.H. Kim and F.L. Lewis, Neural Network Output Feedback Control of overhead crane
Manipulator, IEEE Trans. Neural Networks, Vol.15, 301-309, 1999.
R. Kelly, Global Positioning on overhead crane manipulators via PD control plus a classs
of nonlinear integral actions, IEEE Trans. Automat. Contr., vol.43, No.7, 934-938, 1998.
H.H. Lee, Modeling and control of a three-dimensional overhead crane, Journal of
Dynamic Systems, Measurement, and Control, Vol.120,471-476, 1998.
H.H. Lee, A new motion-planning scheme for overhead cranes with high-speed hoisting,
Journal of Dynamic Systems, Measurement, and Control, Vol.126,359-364, 2004.
E.H. Mamdani, Application of fuzzy algorithms for control of simple dynamic plant, IEE
Proceedings - Control Theory and Applications, Vol.121, No.12, 1585-1588, 1976.
J.A. Mndez, L. Acosta, L. Moreno, S. Torres, and G.N. Marichal, An application of a
neural self-tuning controller to an overhead crane, Neural Computing and Applications,
Vol.8, No.2, 143-150,1999.
K.A. Moustafa and A.M. Ebeid, Nonlinear modeling and control of overhead crane load
sway, Journal of Dynamic Systems, Measurement, and Control, Vol.110, 266-271, 1988.
M.W. Noakes and J.F. Jansen, Generalized input for damped-vibration control of
suspended payloads, Journal of Robotics and Autonomous Systems, Vol.10, No.2, 199205, 1992.
S. Nicosia and A. Tornambe, High-gain observers in the state and parameter estimation of
overhead cranes having elastic joins, System & Control Letters, Vol.13, 331-337, 1989.
21.
22.
23.
24.
25.
26.
27.
11