Intelligent Automation and Soft Computing, Vol. 17, No. X, pp.

1-11, 2011
Copyright 2011, TSI Press
Printed in the USA. All rights reserved

ANTI-SWING CONTROL FOR AN OVERHEAD CRANE WITH FUZZY

COMPENSATION
XIAOOU LI*, WEN YU**
* Departamento de Computacin
** Departamento de Control Automatico
CINVESTAV-IPN
A.P. 14-740, Av.IPN 2508, Mxico D.F., 07360, Mxico

ABSTRACTThis paper proposes a novel anti-swing control strategy for an overhead

crane. The controller includes both position regulation and anti-swing control. Since the
crane model is not exactly known, fuzzy rules are used to compensate friction, gravity as
well as the coupling between position and anti-swing control. A high-gain observer is
introduced to estimate the joint velocities to realize PD control. Real-time experiments
are presented comparing this new stable anti-swing PID control strategy with regular
crane controllers.
Key Words: fuzzy control, anti-swing, overhead crane

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

Intelligent Automation and Soft Computing

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.

2. ANTI-SWING CONTROL FOR THE OVERHEAD CRANE

z
rail
cart

Fx

xw

Fy

R
FR

xw

yw

Fx

yw

Fy

3D Crane

FR

payload
Mcg

(a)

(b)

Figure 1. Overhead crane

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]:

M ( x ) &x& + C ( x, x& ) x& + G ( x ) + F = 1

where

x = [x w , y w , , , R ] , (x w , y w , R )
T

1 = [Fx , Fy , 0, 0, FR ] , Fx
T

rail and along the lift-line,

is

the

(1)

position

of

the

payload,

Fy and FR represent the control forces acting on the cart and

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

Anti-Swing Control for an Overhead Crane with Fuzzy Compensation

dimension of M is 5 5 as in this paper. However, some uncertainties such as friction and

anti-swing control coupling are not included. This overhead crane system shares one important
property with robot systems: the Coriolis matrix C ( x, x& ) is skew-symmetric, i.e., it satisfies the
following relationship [10]

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],

but the swing angles

[ , ]

cannot be controlled using the dynamic model (nm)

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

&x&w << g , &y&w << g , R&& << g

R& << R, & << 1, & << 1

s1 = sin , c1 = cos 1,
The approximated dynamics of

[ , ]

are then

&& + &x&w + g = 0, && + &y&w + g = 0

Since

&x&w =

Fx
Mr

, &y&w =

Fy
Mm

, the dynamics of the swing angles are

&& + 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

, .

The control forces

Fx and Fy are assumed to have the following form

Fx = A1 ( x w , x& w ) + A2 ( , & )
Fy = B1 ( y w , y& w ) + B2 , &

(4)

A1 ( x w , x& w ) and B1 ( y w , y& w ) are position controllers, and A2 ( , & ) and

B2 ( , & ) are anti-swing controllers. Substituting (4) into (3), produces the anti-swing control

where

model

&& + g + MA = MA
&& + g + MB = MB
1

2
r

(5)

Intelligent Automation and Soft Computing

A1
Mr

Now if

and

B1
Mr

A2
Mr

are regarded as disturbances,

and

B2
Mm

as control inputs, then

(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)

k da 2 and k ia 2 are positive constants corresponding to proportional, derivative

and integral gains. Substituting (4) into (1), produces the position control model

M ( x ) &x& + C (x, x& ) x& + G ( x ) + Tx& + D =

]T

= A1 , B1 , 0, 0, FR
where D = A2 , B2 , 0, 0, 0 ,
controller will be designed in next section.

]T .

(7)

Using this model, a position

3. POSITION CONTROL WITH FUZZY COMPENSATION

PD type controller is used for position regulation, which has the following form

= 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

is the desired position, and x&

is the desired joint velocity. Here the regulation problem is discussed, so

A filtered regulation error is defined as

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

rules (Mamdani fuzzy model [16]) is used to approximate it

R i : IF (x w is A11i ) and ( is A21i ) and (x& w is A31i )

and (& is A41i ) and (~
x w is A51i )THEN fx is B1i
IF ( y w is A12i ) and ( is A22i ) and ( y& w is A32i )
and & is A42i and (~y w is A52i )THEN fy is B2i
IF (R is A13i ) and (R& is A23i )
~
and R is A33i THEN fz is B3i

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

Anti-Swing Control for an Overhead Crane with Fuzzy Compensation

mapping from the input vector

x = [xw , y w , , , R ] 5
T

to the output vector

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

is the membership functions of the fuzzy sets

B = 1.
pi

A jip , and w pi is the point at which

(9) can be expressed in matrix form

f = Wt (s )
where

the

parameter

matrix

W = diag W1 ,W2 ,W3 ,

W p = [w p1 L w pl ],

( x ) = [1 , 2 , 3 ] ,

(10)

p = 1p Ll

training algorithms to update the parameter matrix W

].

and

p T

the

data

vector

We will propose stable

and Gaussian function

A .
p
ji

The

position controllers have a PD form with a fuzzy compensator

= [A1 ( xw , x& w ), B1 ( y w , y& w ), 0, 0, FR ]T = Kr f

= K (x x d ) K (x& x& d ) Wt ( s )

]T

where x = x w , y w , , , R ,

x d = xwd , y wd , 0, 0, R d , and x wd ,

(11)

y wd and R d are the

desired positions, K = K p1 > 0. The time-varying weight matrix Wt is determined by the

T

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.

Intelligent Automation and Soft Computing

,

PID anti-swing
control

A2 ,B2

xw* , yw* , R*
PD position A1, B1, FR
control
xw , yw , R

Fx , Fy , FR

Crane

x&w , y& w , R&

High gain
observer

Fuzzy
compensator

xw , yw , R

xw , yw , R

Figure 2. The structure of the anti-swing control with fuzzy compensation

4. PD CONTROL WITH A VELOCITY OBSERVER

The dynamics of the overhead crane (1) can be rewritten in state-space [20]

x&1 = x 2
x& 2 = H 1 ( X , )

(12)

y = x1
where

x1 = [xw , y w , , , R ]

x1 = x , x 2 is the velocity vector,

is the position vector

X = [ x1T , x2T ]T , is the control input. The output is a position measurement,

H1 ( X , ) = M ( x1 ) 1 [C ( x1 , x2 ) x&1 + G ( x1 ) + Fx&1 + ]

(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)

denotes the estimated values of x1 , x 2 respectively;

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

dtd ~z1 = ~z 2 K1~z1

dtd ~z 2 = K 2 ~z1 + 2 H 1

(16)

Anti-Swing Control for an Overhead Crane with Fuzzy Compensation

or in the matrix form:

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

M c = 0.8kg , the mass of payload is M m = 1.3kg ,

I = 0.01kg m 2 , see Figure 1. Here

interfacing is based on a Xilinx FPGA microprocessor, comprising a multifunction analog and

digital I/O board dedicated to real-time data acquisition and control in the Windows XP
environment, mounted in a PC Pentium-III 500 MHz host. The control program operated in
Windows XP with Matlab 6.5/Simulink. All of the controllers employed a sampling frequency of
1kHz .
There are two inputs in the anti-swing model (5), A1 and A2 with A1 from the position
controller and A2 from the anti-swing controller. When the anti-swing control A2 is designed
by (5), A1 is regarded as a disturbance. The chosen parameters of the PID (6) control law were

k pa 2 = 2.5, k da 2 = 18, k ia 2 = 0.01

k pb 2 = 15, k db 2 = 10, k ib 2 = 0.6

Intelligent Automation and Soft Computing

Figure 3. Real-time control for an overhead crane

The position control law in equation (11) is discussed next. In this case there are two types of

]T

input to the position model (nm1), D = A2 , K ,

= [A1 ,L]T .

When the position control

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

Aji = exp (x j m ji ) / 2ji , j = 1L5, i = 1L20

where m ji and

ji

were selected randomly to lie in the interval

(x ) = [ 1 L 20 ] .
T

xwd = 0.5 sin(0.2t ),

(0,1) . Hence,

Wt R 520 ,

The desired gantry position was selected as a circle with

y wd = 0.5 cos(0.2t ) . The resulting gantry positions and angles are

shown in Figure 4 - Figure 6.

It can be seen that the swing angles and are decreased a lot with the anti-swing
controller. From Figure 4 and Figure 5 we see that the improvement is not so good in
direction, because in this direction the inertia comes from the rail, its mass M r is bigger than
the cart M c (

direction). Clearly, PD control with fuzzy compensation can successfully

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.

Anti-Swing Control for an Overhead Crane with Fuzzy Compensation

0.8

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

Figure 4. Positions of PD anti-swing control with fuzzy compensation

0.02

(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

Figure 5. Angles of PD anti-swing control with fuzzy compensation

1

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

Figure 6. Control inputs

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

Intelligent Automation and Soft Computing

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.

Anti-Swing Control for an Overhead Crane with Fuzzy Compensation

21.
22.
23.

24.

25.
26.
27.

11

O. Sawodny, H. Aschemann, and S. Lahres, An automated gantry crane as a large

workspace robot, Control Engineering Practice, Vol.10, No.12, 1323-1338, 2002.
Y. Sakawa and Y. Shindo, Optimal control of container cranes, Automatica, Vol.18, No.3,
257-266, 1982.
W. Singhose, W. Seering, and N. Singer, Residual vibration reduction using vector
diagrams to generate shaped inputs, Journal of Dynamic Systems, Measurement, and
Control, Vol.116, 654 -659,1994.
R. Toxqui, W. Yu, and X. Li, PD control of overhead crane systems with neural
compensation, Advances in Neural Networks -ISNN 2006, Srpinger-Verlgag, Lecture
Notes in Computer Science, LNCS 3972, 1110-1115, 2006.
L.X. Wang, Adaptive Fuzzy Systems and Control, Englewood Cliffs NJ: Prentice-Hall,
1994.
J. Yu, F.L. Lewis, and T. Huang, Nonlinear feedback control of a gantry crane, Proc. 1995
American Control Conference, Seattle, 4310-4315, USA, 1995.
W. Yu and X. Li, Fuzzy identification using fuzzy neural networks with stable learning
algorithms, IEEE Transactions on Fuzzy Systems, Vol.12, No.3, 411-420, 2004.

ABOUT THE AUTHORS

X. Li received the B.S. and the Ph.D. degree in applied mathematics and
electrical engineering from Northeastern University, China, in 1991 and
1995, respectively. Since 2000, she has been a professor in Seccin de
Computacin, Departamento de Ingeniera Elctrica, CINVESTAV-IPN,
Mexico. Her research interests include Petri net theory and application,
neural networks, advanced database systems, computer integrated
manufacturing, and fuzzy system.

W. Yu received the B.S. degree from Tsinghua University, Beijing, China

in 1990 and the M.S. and Ph.D. degrees, both in Electrical Engineering,
from Northeastern University, Shenyang, China, in 1992 and 1995,
respectively. In 1996, he joined CINVESTAV-IPN, Mxico, where he is
currently a professor in the Departamento de Control Automtico. Dr Wen
Yu serves as an associate editor of Neurocomputing, and International
Journal of Modelling, Identification and Control. He is a senior member of
IEEE, and member of Mexican Academy of Science.