The Control System Design of a Gantry Crane Based on H

Control Theory

Ivan Burul, Fetah Koloni, Jadranko Matuko
Department of Electric Machines, Drives and Automation
Faculty of Electrical Engineering and Computing
Unska 3, Zagreb, Croatia
Tel.: +38516129824 Fax: +38516129705
[email protected], [email protected], [email protected]



ABSTRACT This paper describes a robust controller
design procedure based on H

control theory for translational

mechatronic system, trolley with pendulum. The goal is to
achieve fast trolley positioning with minimization of
pendulum swinging. This system, trolley with pendulum,
represents gantry crane. Minimization of the load transfer
time and load swing angle in the crane applications are
conflicting control demands and thus proper control action is
required. However, friction effect is unavoidable in
mechanical parts of the system and significantly influences
the control system performances. In this paper H

theory
based crane controller is designed in order to minimize
friction effect. H

controller has been tested and compared

with the pole placement controller on a laboratory gantry
crane model.


I INTR0B0CTI0N
In modern industrial system, gantry cranes are widely
used for the heavy loads transfer [1][2]. The crane
acceleration, required for the motion, causes an undesirable
load swing having negative consequences on the system
control and safety performances. Beside the load position,
for load swinging minimization it is necessary to control
load swing angle. To achieve more control efficiency,
derivations of the cart position and swing angle should be
controlled as well.
For the anti-sway control of traveling cranes, there are
several solutions, i.e., by fuzzy control, optimal control,
etc. and each of them is reported to be effective [3][4][5],
if the plant model is exactly known. However, in practice,
it is not possible to make the model which accurately
describes the characteristics of the controlled object. In
addition, it is impossible to completely isolate the
controlled object from undesired influences from its
environments. These influences are usually represented as
external disturbances whose effects need to be minimized
by the controller action.
Typical effect that may significantly degrade the
performance of the mechatronic control system is well-
known friction effect. Due to its nonlinear characteristic
friction may prevent the crane control system from
accurate positioning and this problem cannot be solved by
classical approach by adding the integral term to the
controller. Moreover, integral action may cause undesired
nonlinear effect of stick-slip motion [6]. One way to deal
with the friction problem is to generate additional control
in order to compensate friction effect. However, this
approach additionally requires a friction observer to be
designed based on partially or completely known friction
model. The second approach uses high-frequency bias
signal injection. Although it may alleviate friction effects,
it may also excite high-frequency harmonics in the system.
Instead of compensation of the friction effect, in this
paper, H

theory based controller will be designed in order

to make the control system insensitive to the friction effect.
Such the approach generally ensures minimum system
safety and performance, in spite of modeling inaccuracies
and unknown disturbances. In that sense, the friction effect
is considered as an unknown input disturbance whose
effect is to be minimized by the proper control action.
This paper is organized as follows. In section II the main
idea behind H

based controller design is presented. The

gantry crane process is described in section III, while the
anti-swing controller design procedure is presented in
section IV. Experimental verification of the H

based
controller design is given in section V.


II B

BASEB C0NTR0L
H

based control is a modern control approach which

gives a systematic procedure for the design of feedback
control system. It is generally based on minimization of the
H

norm of one or more selected transfer functions. The

H

norm of the transfer function ( ) G s is the peak value of

its frequency response ( ) G je , while for MIMO systems
H

norm is defined as the highest singular value of

corresponding frequency response matrix.

(1)
( )
( ) ( ) max G s G j o e
e
=

(1)
In that sense H

based control system design is basically

frequency domain system design. Therefore, it is necessary
to transform control system specification into requirement
on the shape of the frequency characteristics. Generally,
three transfer functions/matrices play major roles in such
the approach to controller design (defined according Fig.
1):
the open loop transfer function L:
L GK = (2)

the sensitivity function S:

( ) ( )
1 1
S I GK I L

= + = + (3)

the complementary sensitivity function T:

( ) ( )
1 1
T I GK GK I L L

= + = + (4)

183
MIPRO 2010, May 24-28, 2010, Opatija, Croatia
K G
-
r e y u

Fig. 1 Feedback control scheme

The role of these transfer functions can be seen from the
following equations, derived directly from Fig. 2:

( )
GK
y GK r y r Tr
I GK
= = =
+
, (5)

( ) e r y I T r Sr = = = , (6)

in which following identity is obvious:

S T I + = . (7)

Equation (6) shows that the sensitivity function S Sis a
performance indicator: the lower the sensitivity function,
the lower the tracking error and hence, better the
performance. It can be shown that, when also taking into
account model uncertainties, stability in the presence of
multiplicative uncertainty (i.e. robust stability), is obtained
when T satisfies 1
I
T e < , where
I
e is at each
frequency equal to the magnitude of the multiplicative
uncertainty, [7]. So, to obtain a stable closed loop system,
T should be made small in frequency regions where the
model uncertainty is large. This is also already suggested
by equation (5) which indicates that a small T leads to a
small output y (and hence a controller with small gains
which is unlikely to cause instability).
In this way, a trade-off is made between a small
sensitivity function S for good performance and a small
complementary sensitivity function T for robust stability.
The constraint (7) shows that they cannot be made small
simultaneously at the same frequency. Hence, the goal is to
design a controller which has a low sensitivity function S
in the frequency band of interest and a low complementary
sensitivity function T outside this band.


A. Weighting functions as upper bounds on closed-loop
transfer functions

By including weighting functions in the H

control
problem, the designer can shape the closed-loop transfer
functions according to design specification.
Let 1
P
e be a desired upper bound on the magnitude of
S where
P
e is a weighting function specified by the
designer. Note that the requirement:
( ) ( ) 1 , S j j
P
e e e e < (8)
is equivalent to:
( ) ( ) 1, j S j
P
e e e e < (9)
or:
( ) ( ) 1, j S j
P
e e e e <

(10)

These equations show how the specification of an upper
bound on the sensitivity function S leads to a condition on
the H

norm of the weighted sensitivity

P
S e .


B. Shaping different closed-loop functions: mixed
sensitivity

The previous subsection explained how to specify an
upper bound on the sensitivity function S by making use of
the performance weighting function
P
e . However, since
control design is always a trade-off between performance
and robust stability, other requirements need to be taken
into account; for example the requirement that T is small
outside the frequency band of interest. To combine these
requirements on different closed-loop functions, the so-
called mixed sensitivity approach can be applied. In this
approach, specifying an upper bound 1
P
e on S and an
upper bound 1
T
e on T results in the following overall
requirement:
( ) ( ) max 1;
P
T
S
N N j N
T
e
o e
e
e
(
= < =
(

. (11)
After selecting the different closed-loop functions to be
taken into account and their weighting functions, the H

optimal controller is obtained by solving the problem:

( )
min N K
K

, (12)
where K is a stabilizing controller.

G
u
K
w z
y

y
A
u
A

Fig. 3 Standard plant (G) with controller (K) and uncertainty ()

Fig. 3 shows the general framework that will be used in
this article. Any control problem within a linear setting
may be written in this form.
The standard plant G incorporates the plant and provides
an interconnection structure, defining the way in which the
uncertainty block and the controller K act on the system.
The inputs to the standard plant are: the output of the
uncertainty block u

; the disturbances acting on the

standard plant w, such as reference signals and noise; and,
finally, the controller output u.
The outputs of the standard plant are: the input to the
uncertainty block y

; the control objectives z, such as

tracking error and control effort; and the measurable
signals y.
The closed loop transfer function from w to z will be
denoted as T
wz
. The control problem can then be stated as
the minimization over all possible controllers K(s) of T
wz
in
some given norm, for the worst case . This uncertainty
block is allowed to be any arbitrary element of a given
set, representing the difference between the linear model
and reality.
184
To be able to optimize certain design goals (usually
performance), under some constraints, such as robustness
and noise sensitivity reduction, (in fact the weight
functions reflecting the goals which are to be optimized)
the standard plant may depend on a free parameter eR.
By defining such that the demand with respect to one
or more design goals increases when increases, the H

control design problem can be defined as finding the
largest such that there exists a controller K(s) for which :

( )
( ) ( )
, 1, : 1 T s s s
wz
s A A s

. (13)
An important consequence of (13) for the rest of this paper
is that the infinity norm of the uncertainty must be
bounded. Otherwise, no solution can be found [9].
If no uncertainty block is present, a direct computational
solution is available, based on the solution of two algebraic
Ricatti equations. This solution has been implemented in
the software package MATLAB using a small adaptation
of the algorithm given by Glover and Doyle [8].


III BESCRIPTI0N 0F TBE uANTRY CRANE
PR0CESS
Simplified single gantry crane model is depicted in Fig.
4, where x
c
is the cart position, x
p
and y
p
determine
position of the center of the mass of pendulum, M
c
is the
mass of the cart, M
p
is the mass of pendulum, is the sway
angle of pendulum and F
c
is external force acting on the
cart. Zero sway angle, corresponds to a suspended
pendulum vertical rest down position.
p
x
p
y
c
x
k
M
0
c
F >
o
p
M
C
M
0 o >

Fig. 4. Single pendulum gantry crane system


Mathematical equations of the motion can be defined via
Lagrange equations. Nonlinear equations of motion are by
following set of equations [10]:
( )
( )
2
2 2
2 2 2
2
( )
( )
( )
( ) ( )
( )
p p p eq c
c
c p p c p p
p p p p p p c p p
c p p c p p
d
I M l B x t
d dt
x t
dt M M I M M l
d
M l B t I M l F M l g t
dt
M M I M M l
o o
| |
+
|
\ .
=
+ +
| |
+ + +
|
\ .
+
+ +
(14)
( ) ( )
2
2 2
2
( ) ( )
( )
( )
( )
( )
c p p p c p p
c p p c p p
p p eq c c p p
c p p c p p
d
M M M gl t M M B t
d dt
t
dt M M I M M l
d
M l B x t F M l
dt
M M I M M l
o o
o
| |
+ +
|
\ .
=
+ +
| |

|
\ .
+
+ +
(15)




Upon linearization of the nonlinear single gantry crane
model following state space linear model P is obtained:


0 0 1 0 0
0 0 0 1 0
0
1 2 3 7
0
4 5 6 8
x x
c c
U
m
H H H H x x
c c
H H H H
o o
o o
( ( ( (
( ( ( (
( ( ( (
= +
( ( ( (
( ( ( (

, (16)


1 0 0 0
0 1 0 0
x
c
x
c
x
c
o
o
o
(
(
( (
(
=
( (
(

(

, (17)

where:


1.5216 11.6513 0.0049
1 2 3
26.1093 26.8458 0.0841
4 5 6
H H H
H H H
= = =
= = =


and:

1.5304 3.5261
7 8
H H = = .


A. Friction Model

As the uncertainty model is not suitable to incorporate
the friction effect in a standard plant setting [9], an
alternative solution needs to be used. The usual approach
to the reduction of friction influence is the construction a
disturbance observer. In that case an extra state is added to
the controller to be able to estimate the friction force. The
result then resembles the servo compensator approach of
Davison [11] for a constant disturbance input signal, which
uses extra dynamics to describe this signal.
To be able to incorporate the friction phenomenon in the
standard plant setting, but without the necessity of adding
dynamics, the dry friction force is modeled as an external
disturbance force. Hence, disturbance input F
tr
to the
standard plant P is added, as it is shown at Fig. 5.


Iv C0NTR0LLER BESIuN
The H

control system is designed in order to follow a

reference signal and to be robust against friction, as shown
on Fig. 5. It can be seen on the figure there are four
referent signals, for the position of trolley x
r
, for the angle
of pendulum
r
, for the linear trolley speed v
r
and for the
angular velocity
r
. Since the goal of this system to
185
minimize swinging of the pendulum, a reference of the
angle will always have a value of zero. Also, since in
steady state, both linear speed and angular velocity, have
the value of zero, their reference will have a value of zero.
The controller K in that way receives four error signals e,
and provides control signal u for the process P. At the
entrance to the process except the control signal u there is
active and disturbance signal F
tr
which represents the force
of friction.
r
o
r
x
2
e
1
e
u
2
z
1
z
5
z
tr
F
x
o
1 r
W
2 r
W
1
W
2
W
u
W
tr
W
P
K
n
W
n
r
v
r
e
3
e
4
e
v
e
3 r
W
4 r
W
3
z
3
W
4
W
4
z
n
W
n

Fig. 5 H

control system block diagram

Measurement noise is added in the feedback which
signals are measured (position and angle), but the
measurement noise doesnt act on the signal that are
estimated (linear and angular velocity).
System in Fig. 5 is system as it will look like in the end,
when H

optimization calculate controller which will

control the system. But before that, the system should be
described in an open loop (without controller) as:

| |
z v
G
e u
( (
=
( (

(18)
where:
v exogenous input consisting of disturbances,
commands, etc., while the control signal is placed in
a separate vector so it could be distinguished,
e error signals which are forwarded to controller,
z weighted performance variable which may be
arbitrary combinations of state, output, control, error
signals, etc.,
u control signals which are also defined by the system.



System in Fig. 6 has six inputs and five outputs. Signals
are defined as follows:

( )
( )
( )
( )
| |
1
1,1
1
2
2,1
2
, , , ,
3
3 3,1
4
4
4,1
5
x
r
z
P
e x
r
z
P
v e
r
v z z e y u u
e v P
r
z
n e
P
z
F
tr
o
o
e
e
(
(
( (
( (
(
( (
( (
(
( (
( (
( = = = = = = ( (
( (
(
( (
( (
(
( (
(

(
( (


(

, (19)
while open loop system matrix G given by equation (20).
r
o
r
x
2
e
1
e
u
2
z
1
z
5
z
tr
F
x
o
1 r
W
2 r
W
1
W
2
W
u
W
tr
W
P
n
W
n
r
v
r
e
3
e
4
e
v
e
3 r
W
4 r
W
3
z
3
W
4
W
4
z
n
W
n
G
Fig. 6 Open loop H

control system block diagram

A. Selection of weights

Different types of controllers can be achieved by making
changes in weighting functions. To obtain controller,
which will operate quickly, the signal of positioning z
1

should have greater weight. This will of course result in
faster positioning of trolley, but with somewhat greater
sway. There is a possibility to set weight of speed signals,
a linear and angular. Signal z
3
of linear speed is the entire
time held at zero because of his involvement further
complicates the system and its involvement could only
slow down the system. However, if necessary, signal z
3
can
also be included into design. Since the signal of angular
velocity z
4
further helps to keep swinging angle closer to
zero, it was also taken into account.




0 0 0 0
1 1 1 (1,2) (1,1) 1
0 0 0 0
2 2 2 (2,2) (2,1) 2
0 0 0 0
3 3 3 (3,2) (3,1) 3
0 0 0 0
4 4 4 (4,2) (4,1) 4
0 0 0 0 0 0
0 0 0
1 (1,2) (1,1)
0 0 0
2 (2,2) (2,1)
0 0 0 0
3 (3,2) (
Wr W WW P P W
tr
Wr W W W P P W
tr
Wr W W W P P W
tr
Wr W W W P P W
tr
G W
u
Wr W W P P
n tr
Wr W W P P
n tr
Wr W P P
tr




=



3,1)
0 0 0 0
4 (4,2) (4,1)
Wr W P P
tr
(
(
(
(
(
(
(
(
(
(
(
(
(
(
(


(20)


186
The weighting functions W
1
, W
2
, W
3
and W
4
are a
weights related to steady-state characteristics and quick
response; W
u
is a weight related to stability; W
r1
, W
r2
, W
r3

and W
r4
are weights related to a reference input; and W
tr
is
a weight related to the disturbance (friction force) and
following weights are chosen based on the idea mentioned
above.
The weighting functions W
1
is chosen as a frequency
dependent weighting function as:

0,01s+1,408
1
s 0.9918
W =
+
. (21)

The other weighting functions are chosen as constants
since the system is stable (the existence of H

optimal
solution guarantees the system stability) and thus it is
ensured that the steady state error for speeds (linear and
angular), and swinging angle of pendulum are zero.

0, 6
2
0
3
0,1
4
W
W
W
=
=
=
(22)
Weighting function W
u
which determines
complementary sensitivity function T is chosen as:


1.05s 30
s 10000
W
u
+
=
+
. (23)

The series of experiments showed that for this selection
of weighting functions system is the best described when
the weighting functions W
tr
which determines the force of
friction is chosen as:
39 W
tr
= . (24)

Weighting function for the measurement noise W
n
is
chosen as:

3
10 W
n

= . (25)

In this paper reference signals are not weighted, so their
weighting functions are set to value 1.



v EXPERINENTAL RES0LTS
The designed H

controller was experimentally tested on

a laboratory model of the single gantry crane. The
performance of the H

controller was compared to

conventional pole placement controller.
The results, shown in figures 6 and 7 clearly show
advantages of H

controller compared with pole placement

(PP) controller. The rise time t
r
in both cases is practically
the same (about 2.5 s). On the other hand, steady state
error obtained with H

controller is practically zero, while

the pole placement controller has the steady state error of
9%.


Fig. 7 Cart position and pendulum angle


Fig. 8 Pendulum tip position



Fig. 9 Control signal of H

controller



Fig. 10 Control signal of PP controller

187
The next comparison of the two controllers is shown in
Fig. 9 and Fig. 10 which shows a control signal. While
pole placement controller leads to larger control signal
amplitudes to start the pendulum movement, H

controller
produces more dynamic control signal when approaching
the steady state (from t=2s to t=3s). Such the dynamic
control signal prevents the trolley to stop due to static
friction force. In the case of pole placement controller,
control signal in that time period is not large enough to
overcome the static friction force.


vI C0NCL0SI0N
Considering the obtained results, the H

optimization
based controller meets all the requirements. The main goal
of this work, fast trolley positioning with minimization of
pendulum swinging is successfully executed and proved to
be successful in relation with pole placement method.
Comparison of two mentioned methods was done
experimentally and clearly confirmed the superiority of the
H

controller.
H

based controller ensures system response without

steady state error, while this is not possible for the pole
placement method. In H

method steady state error does

not exist because the friction is subsequently taken into
account as an unknown input disturbance although in a
mathematical model the friction is neglected.
Beside this advantage H

method has another advantage

which is possibility of making different types of controllers
and achieving different types of goals by modifying
weighting functions.
It is shown that the influence of friction on the controlled
mechanical system can be largely reduced, within a linear
setting. The standard plant approach has been very useful
for this, in that it allows adding or removing objectives,
simply by adding or removing weight functions. By
altering weight functions one can find a proper trade-off
between various objectives.

REFERENCES
[1] Hanafy M. Omar, Control of Gantry and Tower Cranes,
PhD dissertation Faculty of the Virginia Polytechnic
Institute and State University, USA, 2003.
[2] Tonci Popadic, Fetah Kolonic, Alen Poljugan, A fuzzy
control scheme for the gantry crane position and load
swing control, Faculty of electrical engineering and
computing, Zagreb, 2005.
[3] Michael J. Nalley, Mohamed B. Trabia, Control of
overhead cranes using a fuzzy logic controller, Journal of
intelligent and Fuzzy systems, vol. 8, pp.1-18, 2000.
[4] A. Z. Al-Garni, K. A. F. Moustafa and S. S. A. K. Javeed
Nizami, Optimal control of overhead cranes, Control
Engineering Practice, vol. 3, Issue 9, pp. 1277-1284,1995.
[5] Kunihito Matsuki, Noriyuki Kikuti, Shigeto Ouchi, The
Control System Design of a Traveling Crane using H

Control Theory, Fuji Electric Co., Ltd, Tokyo, Japan.
[6] B. Armstrong, P. Dupont, C. Canudas, A survey of models
analysis tools, and compensation methods for the control of
the machines with friction, Automatica 30 (70) (1994)
10831138.
[7] J.De Cuyper, J. Swevers, M. Verhaegen, P.Sas, H

Feedback Control for signal tracking on a 4 poster test rig
in the Automotive Industry.
[8] Kemin Zhou, Keith Glover, John Doyle, Robust and
Optimal Control, Prentice Hall, 1995.
[9] Gert-Wim van der Linden, Paul F. Lambrechts, Control of
an Experimental Inverted Pendulum with Dry Friction,
IEEE, 0272-1708/93, 1993.
[10] Quanser Inc, Linear Experiment #4: Pole Placement,
Linear Motion Servo Plants: IP02, Quanser, Student
Handout.
[11] E.J. Davison and A. Goldenberg, Robust control of a
general servomechanism problem: The servo
compensator, Automatica vol. 11. pp. 461-471, 1975.



188