A Finite-Element Approach to Control

the End-Point Motion of a Single-Link

Flexible Robot
Eduardo Bay0
Center for Robotics Systems in Microelectronics, Department of Mechanical
Engineering, Universify of California, Santa Barbara, California 93 106
Received July 28, 1986; accepted September 1 1, 1986

A structural finite-element technique based on Bernoulli-Euler beam theory is presented which

will permit the finding of the toques (or forces) that are necessary to apply at one end of a
flexible link to produce a desired motion at the other end. This technique is suitable for the
open loop control of the tip motion. It may also provide a good control law for feedback control.
The finite-element method is used to discretize the equations of motion. This method has a
major advantage in the fact that different material properties and boundary conditions like hubs,
tip loads, changes in cross sections, etc., can be handled in a very simple and straightforward
manner. The resulting differentialequations are integrated via the frequency domain. This allows
for the expansion of the desired end motion into its harmonic components and helps to visualize
the complex wave propagation nature of the problem. The performance of the proposed technique
is illustrated in the solution of a practical example. Results point out the potential that this
technique has in the study of the dynamics and control not only of flexible robots, but also of
any other flexible mechanisms like those used in biomechanics, where high precision at the tip
of very light flexible arms is required.

Journal of Robotic Systems, 4(1 , 63-75 (1987)

0 1987 by John Wiley & Sons, nc. 1 CCC 0471-2223/87/010063-1S04.00
 10974563, 1987, 1, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/rob.4620040106 by Indian Institute Of Technology Bombay, Wiley Online Library on [18/03/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
64 Journal of Robotic Systems-1987

INTRODUCTION
A new generation of faster and more precise robots, like those in use in the mi-
croelectronic industry, is emerging. More and more demand is being placed on robots
with higher speeds, smaller actuators, larger members, and higher payload capacity.
Due to the flexible oscillations, the assumption of rigid dynamics and kinematics is
no longer applicable and the study of these robots considering the flexibility of the
links becomes necessary. Tools for the modeling and analysis of flexible links have
been presented among others by Sunada,’ Chang,’ Geradin ef u L 3 and more recently
by Naganathan and SonL4
On the other hand, the oscillatory behavior of the flexible links makes it difficult
to control the end-point motions for a desired accuracy within an adequate time interval.
The central problem lies in the control of the end-point motion by applying the proper
torques at the actuating points. The interaction between the links’ elasticity and the
control system further complicates the problem, particularly at high speeds. A technique
for end-control motion corroborated by experimental work has been introduced by
Cannon and S ~ h m i t zThey
. ~ proposed a series of stable control strategies for a single-
link flexible manipulator with sensors and actuators situated at opposite ends. A similar
approach based on a more involved analytical model of the link and different sensor
system has been presented by Sakawa et d 6Another method for the suppression of
vibrational motion by using modal space control has been presented by Karkkainen.’
The idea of supplementing torque control at the joints by application of force pulses
at specific points along the links was proposed by Chassiakos and Bekey.’ Lumped
parameter nonlinear continuous models of flexible articulated one-link and two-link
robots, along with control algorithms have been suggested by Nicosia et d9
The method proposed herein aims at controlling the tip motion by directly computing
the torque necessary to apply at one end of the link to achieve the desired motion at
the other end, taking into consideration the flexibility of the link. A variational approach
and the finite element method are used to discretize the equations of motion. The main
advantage of this method lies in the flexible and natural manner in which boundary
conditions and physical properties may be imposed. In particular, conditions commonly
present in robots, such as tip payloads, pointwise inertias like hubs, or changes in
geometric properties and cross sections, may be modeled in a very straightforward
way.
Using the finite-element method the partial differential equations of motion are
transformed into a set of second-order differential equations in terms of the displace-
ment, velocities and accelerations of the finite-element coordinates, and the unknown
forcing function at the end of the link. The solution for the unknown torque requires,
in the time domain, a solution of a set of Volterra-type integrodifferential equations.
In this case the frequency domain provides a simpler way of solving the equations,
since they can be transformed via the fast Fourier transform in a symmetric set of
complex simultaneous equations.
In what follows is the formulation of the problem first described. Hamilton’s principle
provides the variational method upon which the finite-element weak form is built. For
the beam theory the Euler-Bernoulli model is used. Rotary inertia and shear deformation
are thus neglected. The discretized equations of motions are expressed in terms of
 10974563, 1987, 1, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/rob.4620040106 by Indian Institute Of Technology Bombay, Wiley Online Library on [18/03/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
Bayo: End-Point Motion 65

absolute or total displacements. An efficient numerical algorithm is then proposed for

the solution of the equations.
Finally, the proposed technique is applied to a particular flexible one-link robot.
Results illustrate the potential this technique has for the control of tip motions of
flexible mechanisms.

FORMULATION OF THE PROBLEM

Consider the flexible link of constant section depicted in Figure I that has a total
length equal to L, a hub at one end of inertia IH,mass per unit length Z, Young's
modulus E, and constant moment of inertia i. Let us call the total displacement at a
particular location and time v ( x , r ) . This displacement is given by the hub angle O(t)
and the deflection w(x,r) measured from the line x axis so that

Assuming Bernoulli-Euler beam theory and small elastic displacements, and ne-
glecting the internal damping forces (they will be included later on in the formulation),
the total kinetic and potential energy may be expressed in terms of v ( x , t ) as follows:

and

V = d lLEI[v,,(x,r)12 dx. (3)

A dot and subscript x denote partial derivative with respect to time and space, re-
spectively. The work done by the nonconservative actuating torque is equal to

Figure 1. Deformation in a flexible arm.

 10974563, 1987, 1, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/rob.4620040106 by Indian Institute Of Technology Bombay, Wiley Online Library on [18/03/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
66 Journal of Robotic Systems-1 987

The application of Hamilton’s principle yields

l: 6(U - V ) dr + 6: 6W dr = 0.

The substitution of Eqs. (2)-(4) in ( 5 ) renders

After integrating the velocity terms by parts (6) becomes

lr [lL (mv(x,r) 6v + EI v,*, (x,r) 8v,=)dx +

At this point the displacement field can be discretized using a finite element ex-
pansion:

where N i ( x ) are the Hermite polynonials that satisfy the conditions for admissability
and that are defined over the fraction of the length that constitutes the finite element.
v i ( t ) represents the variation of the nodal displacements with time, and N is the total
number of degrees of freedom. The accelerationsand curvatures may now be expressed
as follows:

and

Substitution of Eqs. @)-(lo) in (7) and proper integration will yield the link mass
and stiffness matrices and the forcing vector. The reader is referred to any of the
 10974563, 1987, 1, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/rob.4620040106 by Indian Institute Of Technology Bombay, Wiley Online Library on [18/03/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
Bayo: End-Point Motion 67

numerous works that deal with the finite-element method, e.g., Bathe," for a detailed
explanation of the derivation of the mentioned matrices.
The assembled set of differential equations in terms of the nodal total displacements
and accelerations becomes

The three partitions in Eq. ( 1 1) refer, respectively, to the degree of freedom at the
hub v h = 8, those between the tip and the hub (or internal degrees of freedom) vi,
and the translational degree of freedom at the tip vt, as shown in Figure 2. The physical
property matrices, mass, stiffness, and damping, are partitioned correspondingly. The
internal viscous damping matrix can now be formed directly from the mass and stiffness
matrices for a given damping ratio in terms of the Caughey series." Equation (11)
may be expressed in reduced form as follows:

M [I [ ] I:[
Vi +C Ci +K vi

The total response of the system is usually contained in only a few modes of vibration.
It may be convenient at this stage to reduce the set of Eqs. (12) by using the normal
modes of the system. A modal analysis followed by a change to generalized modal
coordinates with the corresponding matrix transformations will render a reduced set
of uncoupled equations.
This reduction is avoided here for the sake of generality and also to maintain the
formulation as clear as possible without further analytical complications. However,
the modal approach to reduce the total number of degrees of freedom may become
necessary if this analytical technique is to be used for control purposes. It is worth
pointing out the major advantage introduced using finite elements, by which mode
shapes and frequencies can readily be obtained when complicated boundary conditions

Figure 2. Finite-element degrees of freedom.

 10974563, 1987, 1, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/rob.4620040106 by Indian Institute Of Technology Bombay, Wiley Online Library on [18/03/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
68 Journal of Robotic Systems-1 987

are present. The solution of the eigenvalue problem directly from the partial differential
equation and boundary conditions can be very involved, particularly when the latter
are complex.
We are seeking the solution for T( t) in the set of differential equations (12) that
produces a desired motion at the tip v , ( t ) . In the time domain T(r) could be expressed
as convolution integrals between appropriate impulse responses at the tip, leading to
a set of Volterra type of integrodifferential equations. It is more convenient and effective
to solve for T ( t ) by a frequency-domain analysis.
The system of equations (12) can be transformed by means of the fast Fourier
transform (FFT)into independent sets of simultaneous complex equations. The number
of sets is equal to the number of Fourier pairs, and the frequencies associated with
each pair are equal to

where P is the total time over which the response is analyzed. The set of complex
equations for a particular frequency W becomes

[M+z
1 C - r K
W2 I[;] vi = “I.
where the caret stands for the Fourier transform. In the frequency domain the damping
forces may be expressed in a more convenient way by introducing the complex stiffness
matrix.I2 The set of equations (14) may be thus expressed as

[M - &K*][:]

v,
[
I f(8W) 1,
where

K* = K[l + 2i€J, (16)

and where 6 represents the material damping ratio. Equation (15) may be expressed
in simplified notation as

.[I=[ ;I. f(W)

The transfer matrix H is a complex regular symmetric matrix except for W = 0 for
which it is indeterminant. However, the system does not need to be analyzed for
-
H) = 0 because the zero frequency content of the desired tip acceleration is also zero.
 10974563, 1987, 1, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/rob.4620040106 by Indian Institute Of Technology Bombay, Wiley Online Library on [18/03/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
Bayo: End-Point Motion 69

f ( W ) may be obtained by solving the system of equations (17) for each of the fre-
quencies W.It is important to note that each W also represents the frequency of the
harmonic waves that integrate the given motion. Those flexural waves having fre-
quencies similar to the natural frequencies of the system will be amplified due to
resonance effects, and will form the main part of the total response. The very high-
frequency waves will be damped out quickly by the internal damping mechanisms. It
becomes obvious then that previous knowledge of the natural frequencies of the system
will help in identifying that part of the frequency spectrum in which the total response
is concentrated. Equation (17) will need to be solved for only those frequencies
contained in that part of the frequency spectrum.
The solution of Eq. (17) yields

where G is the inverse of the transfer matrix H.

From (1 8) it is obvious that

and therefore

!herefore the torque may be computed directly from the given tip acceleration. Once
T(E) is known, the angle at the hub and the internal displacements are given by

At this stage it is important to emphasize the fact that the whole matrix G need not
be calculated. As shown above, only the first column is necessary to obtain both T
and the response of the rest of the flexible link. A lot of computational effort may be
saved by this fact, since once the matrix H has been triangularized only one forward
reduction and backsubstitution have to be performed to calculate the first column of
G.
The final results in the time domain may be obtained through the application of the
inverse fast Fourier transform.

APPLICATION
The formulation explained above is applied to a flexible link like that shown in
Figure 1, with the following characteristics:
 10974563, 1987, 1, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/rob.4620040106 by Indian Institute Of Technology Bombay, Wiley Online Library on [18/03/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
70 Journal of Robotic Systems-1 987

length = 150cm,
damping ratio = 0.02,
E = 2 X lo7 N/cm2,
p = 0.008kg/cm3,
I = 0.25 x 10-2cm4,
A = 1 cm2,
I , = 1OOOkg x cm2.

The link is divided into 10 finite elements, giving a total of 21 degrees of freedom.
The first element is considered 10 times more rigid than the rest to account for the
stiffness of the hub.
The imposed tip acceleration and displacement are shown in Figures 3 and 4,
respectively. After the finite element consistent mass and stiffness matrices are formed,
a modal analysis gives the following first periods of vibration:

To = (rigid body motion),

Q)

Tz = 0.525 SW,
T2 = 0.182 sec,
T3 = 0.0132 sec,
T4 = 0.0375 SW,

Ts = 0.0220 sec.

np m E m u n o N
CM/SEC-P

2oo r
150 -
100 -

50 -

-_

-
0 - _.

-50 -
-100 -
-150 -
1 1 1 I
0.25 0.50 0.75 1 .oo 1.25 I 30
TIME (SEC)

Figure 3. Imposed tip acceleration.

 10974563, 1987, 1, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/rob.4620040106 by Indian Institute Of Technology Bombay, Wiley Online Library on [18/03/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
Bayo: End-Point Motlon 71

TIF DISPUCEMENT
CM
11 -
10
9
t 1
/,-------
/'

7-
6-
5 -

4-
3-
2-
1-
1 I I I I

Figure 4. Imposed tip displacement.

The torque that needs to be applied at the hub to produce the desired tip motion is
calculated according to the procedure explained above, and it is shown in Figure 5 .
This figure also illustrates the torque needed to produce the same tip motion on an
infinitely rigid link having the same mass as that of the flexible link. The peak value
of the torque for the flexible link is half of that required in the rigid link. It is worth

-00 - I
I
I
I
-100 - '1I --I]I I
I I I
 10974563, 1987, 1, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/rob.4620040106 by Indian Institute Of Technology Bombay, Wiley Online Library on [18/03/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
72 Journal of Robotic Systems-1 987

pointing out that when a direct dynamic analysis is performed and the system is analyzed
with the calculated torque as input, tip displacement, and accelerations are obtained
which exactly coincide with the desired ones.
These results also show that in order to get the imposed tip motion of the flexible
link the torque needs to be applied 0.3 sec before the actual tip displacement starts
taking place. This is due to the fact that waves along the link move at a finite speed
that depends on the arm flexibility and frequency of excitation. This finite wave speed
introduces a time lag between actuator and tip response. The torque keeps on actuating
another 0.3 sec after the tip has reached its final position, during this time the torque
is absorbing or compensating the reflected waves that come from the tip. Figure 6
compares the hub rotation of the flexible link to that of the rigid link.
A second example is run with the same link, but with its moment of inertia modified
to I = I t z cm4;that is, four times more rigid than the previous one. This configuration
may be considered more rigid than flexible. The torque needed to produce the desired
motion in this case is plotted against that needed in the case of an infinitely rigid link
in Figure 7. As can be seen, the calculated torque is more equal to the rigid link
torque; furthermore, the time lag between actuating time and response is smaller than
in the previous case. This is because of the increase of wave velocity due to the
corresponding increase of rigidity.
Figures 8 and 9 show the oscillations of the tip motion when the torque that produces
the desired tip motion in a rigid link is applied to the flexible links of inertia I = f
l t 2 cm4 and I = 1C2cm4, respectively. The tip reaches the desired end position
only after several seconds in which the tip oscillates near that position. Figures 8 and
9 also show how after 4.5 sec the tip is still oscillating heavily in the case of I = f

HUB ROTATION
md
0.07000

’
0.06000

0.05000
/
/
/-

-7-
0.04000

0.03000

0.02000

p.01000

‘0.00000

- o . o l o ~-
~oo -La
0.25 0.50 0.75 1.00 1.25 1.so
TIME (SEC)

Figure 6. Rotations at the hub. Flexible link 1 = 4 1Cr2 cm4 (solid line) and rigid link (dashed
line).
10974563, 1987, 1, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/rob.4620040106 by Indian Institute Of Technology Bombay, Wiley Online Library on [18/03/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
73

4 10-* cm4
cm4 (solid line) and the rigid link
i

Figure 8. Tip displacements produced by rigid-link torque in the flexible link I =

I

= 10-’
Figure 7. Calculated torques for the flexible link I

TIME (SEC)

(solid line) and in the rigid link (dashed line).

I

n p DISPLACEMENT
Bayo: End-Point Motion

I
-

-80 -
-
-
-40 -

-
-
-

-60 -
80 -

40-

CM
120
100

60

-1 20
20

(dashed line).
0

-100
-20
 10974563, 1987, 1, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/rob.4620040106 by Indian Institute Of Technology Bombay, Wiley Online Library on [18/03/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
74 Journal of Robotic Systems-1 987

n p DISPLACEMENT
SM
12 -
10 -I
i
i
at-

2t- I i
1W2cm4 and slightly for I = 1W2cm4. This waiting time is much larger than that
needed for the computed torque to compensate the reflecting waves in the flexible
links.

CONCLUSIONS
A method has been presented for the calculation of the end torque that is required
to produce a desired motion at the tip of a flexible link. This method is based on a
finite-element frequency-domain formulation for the solution of the equations of mo-
tions that define the flexiblebehavior of the link. The computed torque, which is much
smaller than that required for a rigid link of the same weight, reproduces exactly the
desired tip motion without overshoot. The implicit assumption of a linear response in
the system is necessary for the frequency domain analysis to be applicable.
The expansion of the desired end motion into its harmonic components helps to
understand the wave-propagation nature of the problem, by which the different waves
generated by the torque at one end propagate along the link and subsequently reflect
back at the other end to return to their original position. The applied example dem-
onstrates the accuracy of the method.
The proposed technique is suitable for open-loop control, and it may provide a good
closed-loop control law. Further work is necessary to increase the numerical efficiency
of the method and to extend it to multilink robots. These are the subjects of future
work that the author will undertake.

The author would like to acknowledge the support of this work by the National Science
Foundation, Grant No. 8421415, through the Center for Robotics Systems in Microelectronics.
 10974563, 1987, 1, Downloaded from https://onlinelibrary.wiley.com/doi/10.1002/rob.4620040106 by Indian Institute Of Technology Bombay, Wiley Online Library on [18/03/2025]. See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions) on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License
Bayo: End-Point Motion 75

References
1. W. H. Sunada and S. Dubowsky, “On the Dynamic Analysis and Behaviour of Industrial
Robotic Manipulators with Elastic Members,” ASME J . Mechun. Transmiss. Autom.
Design, 105, 42-50 (1983).
2. L. W.Chang, “Dynamic Analysis of Robotic Manipulators with Flexible Links,” Ph.D.
thesis, Purdue University, 1984.
3. M. Geradin, G. Robert, and G. Bemardin, “Dynamic Modelling of Manipulators with
Flexible Members,” in: Advanced Software in Robotics, A. Dantline and M. Geradin
(Us.), Elsevier, North-Holland, 1984, pp. 27-39.
4. G. Naganathan and A. H. Soni, “Non-Linear Flexibility Studies for Spatial Manipulators,”
in: Proceedings of the 1986 IEEE International Conference on Robotics and Automation,
San Francisco, April 1986, pp. 373-378.
5 . R. Cannon and E. Schmitz, “Initial Experiments on End-Point Control of a Flexible One-
Link Robot,” lnt. ./. Robotics Res. 3,(3), 62-75 (1984).
6. I. Sakawa, F. Matsuno, and S.Fukushima, “Modelling and Feedback Control of a Flexible
Arm,” J. Robotic Syst., 2, 453472 (1985).
7. P. Karkkainen, “Compensation Manipulator Flexibility Effects by Modal Space Tech-
niques,” in: Proceedings of the 1985 IEEE International Conference on Robotics and
Automation, St. Louis, March 1985, pp. 972-977.
8. A. G. Chassiakos and G. A. Bekey, “Pointwise Control of a Flexible Manipulator Arm,”
in: SYROCO ‘85. Barcelona, November 1985, pp. 113-117.
9. S. Nicosia, P. Tomei, and A. Tomambe, “Dynamic Modelling of Flexible Robot Ma-
nipulators,” in: Proceedings of the 1986 IEEE International Conference on Robotics and
Automation. San Francisco, April 1986, pp. 365-372.
10. K. J. Bathe, Finite Element Procedures in Engineering Analysis, Prentice-Hall, Englewood
Cliffs, NJ, 1982, Chap. 5 .
11. T. K. Caughey, “Classical Normal Modes in Damped Linear Systems,”./. Appl. Mechun.,
27, 269-271 (1960).
12. L. Meirovitch, Analytical Methods in Vibrations, Macmillan, New York, 1967, Chap. 9.