Copyright © IFAC Algorithms and Architectures

for Real-Time ControL SeouL Korea, 1992

DESIGN OF A SIMPLE AND ROBUST CONTROL

SYSTEM FOR A SINGLE-LINK
FLEXIBLE ROBOT ARM

Kang-Bark Park, Ju Jang Lee

Department of Electrical Engineering, Korea Advanced Institute of

Science and Technology , 373-] Kusong-dong Yusong-gu 305-70] , Korea

Abstract. Lightweight robot manipulators have considerable structural flexibility.

Hence, the elastic behavior of the robot arms must be considered in control system
design. Owing to the complexity of a dynamic model of flexible robots, it is desir-
able to design a controller simple. Furthermore, since the robot must handle a wide
variety of payloads, the robustness of the control system becomes very important. In
this paper, a simple and robust control system is proposed. The closed-loop system is
shown to be asymptotically stable in the sense of Lyapunov and robust to uncertainty
in system parameters. The simulation results are presented to show the robustness
of the simple controller against payload variations.

Keywords. Variable structure control; Control system design; Robust Control; Feed-
back control; Stability; Flexible manipulator.

INTRODUCTION ported, and so on. For the flexible manipulators,

Robot manipulators have been widely used both
in dangerous circumstances and for industrial ap-
plications. They have been traditionally modeled
as a chain of rigid links with colocated actua-
tors and sensors, to ensure stable and reliable
control. Moreover, the robot arms are gener-
ally made large and massive in order to remain
its rigidity while carrying an assigned payload.
Present generation manipulators are limited to a
load-carrying capacity of typically 5-10% of their
own weight by the need of structural rigidity.
Today, there is an increasing requirement for
manipulators with high speed, precision and
payload-handling capabilities as a result of de-
mand for higher productivity. These qualities
cannot be achieved with existing massive and
heavy robot manipulators. For higher operat-
ing speeds, the manipulators should be made
lightweight, but lighter members are more likely
to deform elastically. Therefore, it is necessary
to include dynamic effects of the distributed link
flexibility in the model of the manipulators.
Such flexible manipulators provide diverse ad-
vantages as follows: higher speed, smaller ac-
tuators, lower energy consumption, lower over-
all cost, safer operation due to reduced inertia,
less bulky design, lighter overall mass to be trans-
however, the equations describing the manipula-
tor dynamics become more complex; besides, they
complicate the control system design which fo-
cuses primarily on the compensation for bending
effects. Furthermore, since the robot must handle
a wide variety of pay loads, the robustness of the
control system becomes very important.
Recently, numerous investigators have analyzed
several modeling methods of a flexible manipu-
lator. One of them is a modal analysis method .
Dominant eigen-modes are identified by experi-
ment, approximated modes are obtained through
solving the governing partial differential equa-
tions and assumed modes can be chosen from ap-
proximation of the actual system dynamics. Fi-
nite element method (FEM) has also been used.
This method discretizes the actual system into
a number of elements, whose elastic and inertia.l
properties a.re obtained from the actual system.
These methods give the approximated static and
dynamic properties of the actual system. Be-
cause the dynamics of mechanical systems with
distributed flexibility are described by infinite-
dimensional mathematical models, and a finite-
dimensional model of the system is needed for
the design of a finite-dimensional controller.
There are also various schemes proposed for the
design of controllers. A technique for end-point

181
control corroborated by experimental work has
been introduced by Cannon and Schmitz (1984).
By using the Linear Quadratic Gaussian (LQG)
controller design method, they successfully im-
plement an noncolocated controller for the robot.
However, since the controller is sensitive to pa-
rameter variations, its performance will be de-
graded when payload or typical parameters of the
robot are varying with time. Rovner and Can-
non (1987) used Recursive Least Square (RLS)
algorithm to identify the system transfer function
with unknown payload on the tip . The scheme
requires a learning period which takes about two
seconds with a sampling frequency of 50Hz. The
experiments have shown good results, but the ro-
bustness of the algorithm still depends upon the
number of coefficients of the transfer function to
be identified. Other approaches are as follows: fi-
nite element approach, modal control, Model Ref-
erence Adaptive Control (MRAC), conventional
ertia of the beam about the motor armature ( =
~pL3 ).
The following assumptions are made:
• The deflection w is small ( w « L ).
• Euler-Bernoulli beam assumptions are used.
• Beam inertia and flexibility are uIliformly
distributed over the link length.
• The motion occurs only in the horizontal
plane.
Because of the last assumption, the effects of
gravity are not considered.
The displacement of points along the deformed
profile of the beam is described in terms of radial
and circumferential coordinates x and y. From
Fig. 1, it is apparent that y is related to the an-
gle of rotation of rigid mode, qo' and the flexural
displacement of the beam , w, as follows :
y( x, t) = w (x, t) + x' q (t) .
0

control, feedforward control, and combined state

space and frequency domain techniques, accelera- In terms of this variable a fourth-order partial
tion feedback control (Balas, 1978; Book and Ma- differential equation of motion for a single-link
jette, 1983). flexible arm is given in the form (Cannon, 1984)

For the design of a flexible manipulator controller, (1)

there are two facts to consider: first, the simplic-
ity; second, the robustness. Owing to the com- with four boundary conditions:
plexity of dynamic model of flexible manipula-
tors, it is desirable to design a controller simple. y(O, t) = 0, (2)
Furthermore, robustness of the controller is also EIy"(O, t) + T - hij" 0, (3)
very important for the robot to handle a wide
EIy"(L, t) 0, (4)
range of payload variations. In general, control
systems with state observer have been used. The
EIy"I(L , t) - Mpy(L, t) 0, (5)
control systems, however, require heavy compu- where' and I denote the derivative with 'respect
tational load. Moreover, observer must be var- to time and x . Then the following equation can
ied in operation to cope with payload variations. be obtained.
Hence, load forecast is needed.

In this paper, a simple and robust controller is

proposed . For the simplicity, the control system where IT = h +h+MpL2, qo is a modal function,
only uses measurable data such as : hub ratio, 4>.(x) is a mode shape, and i = 0, 1,2, " '. The
tip position and tip ratio. Sliding mode control model can be reconfigured in the state-space form
method, one of robust control methods, is intro-
duced for the robustness. On account of robust- Ax+BT, (6)
ness, load forecast is not necessary. y ex,
where
DYNAMIC MODEL OF A
FLEXIBLE ROBOT ARM = [qo cio q1 ql ' " q" q"

Consider a uniform, slender beam connected via °1

a rigid hub to the armature of an electric mo-
tor and having a payload as shown in Fig. 1. °° 1
Where x, L, p, E, I and Mp are distance along
the length of the beam, length of the beam, mass
A
°
-w~ -26wl

per unit length, Young's modulus of elasticity and

cross-sectional area moment of inertia and pay-
load mass. h is the hub inertia and h is the in-

182
Bt = I~ [0 1 0 4>~(0) ... 0 4>~(0)] Srigid erigid + '\1 et
-iio+ '\1 et
L 0 4>dL) 0 ... 4>n(L) 0
C O L 0 4>dL)· · · 0 4>n(L)
1 1
IT
\.
- - T + Alet

[ = o.
o 1 0 4>~(0)··· 0 <p~(0)

Therefore, equivalent iuput torque is

where t denotes transpose and the three rows of
C are the tip-position sensor, tip-rate sensor and
Teq = iT'\let.
the hub-rate sensor measurement vectors.
Now, consider the following control law
Practically, a model based on as few as four
modes may be suitable for simulation and control
system design. In this paper, the model based on
three modes ( n = 3 ) is considered. where '\1 > 0, '\2 > 0, eh = ed - e,,, eh denotes
the hub angle, and

DESIGN OF A CONTROLLER
I;' yITm .. ITm;",
s et + '\let,
In this paper, one of the control objectives is to in-
troduce additional damping into the flexible mo-
tion . This has been done by using additional sen-
sors to measure the flexible link vibrations and
sgn(s)

K =
{-~ s>O
s=O
s < 0,
1(.8 - l)'\lft - '\2fhl + f3M + '1,
feed measurements of sensors back to the con-
M > 0,
troller. Generally, strain gauges have been used
in order to observe the states. This method, how-
ever, leads to heavy computational load. Hence,
a simple control system using only directly mea-
'1 > O.
surable outputs is presented in this section.
With this controller, following lemma is obtained.
Theoretically, any flexible system has an infinite
number of elastic modes. Due to physicallimita- lemma 1 For the system (6) and the proposed
tions, a limited number of sensors and actuators controller (8), the sliding mode exists provided
can be applied, thus restricting the controller de- that
sign to a few critical modes in case of state utiliza-
tion. Note that outputs of sensors would contain
information about the unmodeled as well as the
modeled modes. This is referred to as observa-
sup
t
If
i=l
I
<pi1L) ii;(t) < M.

tion spillover. Similarly, the control action would

proof
affect both the modeled and unmodeled modes
leading to the control spillover. Based on the
It is sufficient to show that ss ~ 0, and ss = 0 if
and only if s = o.
work done by Balas (1978), the higher unmod-
eled modes may have a harmful or destabilizing
effect on the system response . To avoid the ob-
servation spillover, low pass filtering of the sensor
outputs can be used. Control spillover can also be
avoided by inserting the Low Pass Filter (LPF)
in the controller output stage . In this paper, the
controller is designed based on the model equa-
tion (6) .

For the system (6), consider the following switch-

ing surface

(7)

where erigid = ed - qo' et = ed - et, ed denotes the

desired tip angular position and et denotes the tip
angular position . In the sliding mode, Srigid = O.
Applying this condition to equation (7) leads to
the following equation:

183
 Now, consider the meaning of the term )..2eh ( =
s~~ [(~: -1)>'let->'2fn-KS9n(S) hub error rate) in the controller (8). Assume
that the exact value of IT is known and K is zero
- :T f L
rP;i ) q;(t)] for the simplicity of analysis. Then the control
law is rewritten in the form:
IT ;=1

< -rllsl· o (9)

Furthermore, following theorem is also obtained And the system (6) is rewritten as follows:
for the stability.
theorem 1 For the system (6) and the proposed
controller (8), the overall system is asymptotically
stable in the sliding mode. Inserting equation (9) into this equation gives:

Proof
Let us choose the Lyapunov function candidate
as

Then V is a p.d.f. And in the sliding mode,

In this equation, the right-hand side does not con-

tain the ith mode and the sign of rP~(O)rP;(L) are
Therefore,
not same for all modes. Therefore, it is desir-
n able to use the eh factor in the feedback mech-
V L [w;q;q; + q; (- w;q; - 2~;w;q; anism. Since the factor )..2eh leads to the term
;=0 >'2 (1,6:(0))2 rl;, additional damping is introduced.
Because dominant vibrational mode is generally
the first one and there is an uncertainty in pay-
load, design of )..2 value must be based on these
factors.

SIMULATION RESULTS

In order to demonstrate the effectiveness of the

controller developed in the previous section, nu-
merical simulation has been performed on a
single-link three-mode flexible manipulator.
;=0
TABLE 1 Modal Parameters ( 0.0 kg Payload)

i w; ( rad/sec ) rP~ (0) 1,6; (L) e;

0 0 1 1 0
Therefore, V = 0 if and only if q; == 0 V i . This 1 42.0 5.478 -0.359 0.0015
fact implies that e't = ei. = 0 and qi = 0 V i and 2 108.2 5.929 0.217 0.002
q; = 0 V i ~ 1. Furthermore, s = et + et = O. 3 261.0 2.387 -0.178 0.002
Hence, et = 0 and for the rigid mode,

Od - t
;=0
L
rP;i ) q; TABLE 2 Modal Parameters ( 0.2 kg Payload )

Od - qo i w; ( rad/sec ) rPHO) 1,6; (L) ~;

o. 0 0 1 1 0
1 41.1 6 .731 -0.286 0.0015
Therefore, qo = Od, q; = 0 Vi ~ 1 point in the 107.4 7.367 0.167 0 .002
2
state space is asymptotically stable by La Salle's 259.7 2.965 -0.135 0.002
3
theorem (Vidyasagar, 1978). 0

184
TABLE 3 Modal Parameters ( 0.4 kg Payload ) tional modes of the system as states . Hence, it re-
quires heavy computational load because of many
matrix and vector operation. In addition, be-
i Wi ( rad/sec ) ~(O) ifJi (L) ei cause the conversion matrix that is used in state
0 0 1 1 0 estimation is affected by payload variation, con-
1 40.6 7.788 -0.244 0.0015 ventional control system may be sensitive to pay-
2 107.1 8.565 0.140 0.002 load variation and parameter uncertainty. In the
3 259.1 3.446 -0.113 0.002 proposed controller, these drawbacks are elimi-
nated by using robust control law: variable struc-
The model parameters for the first three flexible ture controller. It is clear that the proposed con-
modes with various pay loads are given in TABLE troller has a simple form. In order to verify the
robustness of the controller, the simulation results
1,2 and 3.
are given in the previous section.
In the TABLE 1, 2 and 3, mode number zero
means the rigid mode. For the rigid mode, 4>0 (x)
is defined as follows: REFERENCES

Balas, M. J. (1978). Feedback control of flexible

From these tables, it is obvious that the system
parameters are largely affected by payload varia-
tions .
It is assumed that the flexible manipulator is ini-
tially at rest . The controller in equation (8) is
discontinuous and it is well known that synthe-
sis of such a controller gives rise to chattering of
trajectory about sliding surface s = O. In order
to avoid the chattering phenomenon, the function
sgn(s) in the controller (8) has been replaced by
sat (s) . The function sat (s) is defined as follows
s> 6
sat( s) Isl ::;
6
s < -6 ,
systems. IEEE Trans. Aut. Control., 23,
673-679.
Book, W . J., and M. Majette (1983) . Controller
design for flexible, distributed parame-
ter mechanical arms via combined state
space and frequency domain techniques.
J. Dynamic Syst., Meas, and Contr., 105,
245-254.
Cannon, R. H. Jr., and E. Schmitz (1984).
Initial experiments on the end point
control of a flexible one-link robot.
Int. J. Robotics Res., !t 62-75.
Itkis, U. (1976). (Ed .) Control Systems of Vari-
able Structure. Wiley & Sons, New York.

where 6 > O. Meirovitch, L. (1967). (Ed.) Analytical Methods

in Vibrations. MacMillan, New York.
Figs. 2~5 show the performance of the proposed
Rovner, D., and R. H. Jr, Cannon (1987). Exper-
controller. These figures demonstrate that the
iments toward on-line identification and
tip angular position Bt is successfully regulated
control of a very flexible one-link manip-
for various payloads. Figure 3 shows that there
ulator. Int. J. Robotics Res., ~ 3-19.
is no chattering in the control torque T .
Utkin, V. I. (1977) . Variable structure systems
In practice, gripper of the flexible manipulator with sliding modes. IEEE Trans. Aut.
may drop payload in the course of motion . In Control., 22, 212-222.
this case, the payload is changed abruptly. Figs.
4~5 show the performance of such a case. In Fig. Vidyasagar, M. (1978). (Ed .) Nonlinear Systems
4, gripper drop the O.4kg payload at 0 .9sec, and Analysis. Prentice-Hall, New Jersey.
Fig . 5 shows the control input .

CONCLUSION

A simple and robust control system of a single-

link flexible arm is proposed. This controller
does not need to estimate the modal functions
of the system. It only uses output measurements
such as: tip position, tip position rate and hub
rate. Conventional control system uses vibra-

185
 Regulation - Payload is varied from O.4kg 10 O. ~ at 0.9 aocond

0..9

y 0..8

0..7
XI
~ 0..6
8 0.5
1 0..4
~
0..3

0..2

0..1

Fig . 1. Geometry of the flexible arm. Time (IIOC)

Fig. 4. Tip position ; payload variation .

Regulation - Pay load is varied from O.4kg to O.~ at 0.9 KlCond

Time (aoc) TUne (aec)

Fig. 2. Tip position. Fig . 5. Control Input; payload variation .

Regulation fo< o..o.kg. 0..2kg. 0..4kg poyl",,"

-2

Time (ICe)

Fig. 3. Control input torque .

186