Sei sulla pagina 1di 27

http://jvc.sagepub.

com
Journal of Vibration and Control
DOI: 10.1177/1077546304042058
2004; 10; 1415 Journal of Vibration and Control
Anthony Green and Jurek Z. Sasiadek
Dynamics and Trajectory Tracking Control of a Two-Link Robot Manipulator
http://jvc.sagepub.com/cgi/content/abstract/10/10/1415
The online version of this article can be found at:
Published by:
http://www.sagepublications.com
can be found at: Journal of Vibration and Control Additional services and information for
http://jvc.sagepub.com/cgi/alerts Email Alerts:
http://jvc.sagepub.com/subscriptions Subscriptions:
http://www.sagepub.com/journalsReprints.nav Reprints:
http://www.sagepub.com/journalsPermissions.nav Permissions:
2004 SAGE Publications. All rights reserved. Not for commercial use or unauthorized distribution.
at PENNSYLVANIA STATE UNIV on February 5, 2008 http://jvc.sagepub.com Downloaded from
Dynamics and Trajectory Tracking Control of a
Two-Link Robot Manipulator
ANTHONY GREEN
JUREK Z. SASIADEK
Department of Mechanical and Aerospace Engineering, Carleton University,1125 Colonel By
Drive, Ottawa, Ontario, K1S 5B6, Canada
(Received 15 September 20021 accepted 23 May 2003)
Abstract: Operational problems with robot manipulators in space relate to several factors, most importantly,
structural flexibility and subsequent difficulties with their position control. In this paper we present con-
trol methods for endpoint tracking of a 12.6 12.6 m
2
trajectory by a two-link robot manipulator. Initially,
a manipulator with rigid links is modeled using inverse dynamics, a linear quadratic regulator and fuzzy
logic schemes actuated by a Jacobian transpose control law computed using dominant cantilever and pinned
pinned assumed mode frequencies. The inverse dynamics model is pursued further to study a manipulator
with flexible links where nonlinear rigid-link dynamics are coupled with dominant assumed modes for can-
tilever and pinnedpinned beams. A time delay in the feedback control loop represents elastic wave travel
time along the links to generate non-minimum phase response. A time delay acting on control commands
ameliorates non-minimum phase response. Finally, a fuzzy logic system outputs a variable to adapt the con-
trol law in response to elastic deformation inputs. Results show greater endpoint position control accuracy
using a flexible inverse dynamics robot model combined with a fuzzy logic adapted control law and time
delays than could be obtained for the rigid dynamics models.
Key Words: Robot control, flexible dynamics, assumed modes, fuzzy logic
NOMENCLATURE
A state matrix
B control matrix
C state measurement connection matrix
c wave velocity (m s
1
) (value 288.33)
c
i
link centroid
D direct transmission matrix
d simulation delay time (ss)
E Young modulus (Pa)
EI flexural rigidity (N-m
2
)
e
x
, e
y
position errors
e
x
, e
y
velocity errors
F
i
generalized force
Journal of Vibration and Control, 10: 14151440, 2004 DOI: 10.1177/1077546304042058
1 2004 Sage Publications
2004 SAGE Publications. All rights reserved. Not for commercial use or unauthorized distribution.
at PENNSYLVANIA STATE UNIV on February 5, 2008 http://jvc.sagepub.com Downloaded from
1416 A. GREEN and J. Z. SASIADEK
G dynamic coupling matrix
g gravitational acceleration vector (m s
12
) (value 9.81)
Hz frequency (cycle s
11
) (value 1.0)
J123 Jacobian of direct kinematics
J linear quadratic regulator (LQR) performance index
J energy (Joule)
K stiffness matrix
K
d
derivative gain matrix
K
p
proportional gain matrix
K
i
stiffness constant for ith link
K
N
normalizing gain
kg kilogram
1 Lagrangian function
L link length (m) (value 4.5)
M inertia matrix
M
kp
k,p element of M
M
i
generalized distributed mass for ith link
m meter
m
i
total mass of ith link (kg) (value 1.5075)
m1x3 mass as a function of distance x along link
N force (Newton)
Pa pressure, stress (Pascal)
p1x4 t 3 distributed load
Q LQR weighting matrix
q
i
generalized coordinates
R LQR weighting matrix
r
04Ci
link centroid position vector
s second
ss simulation step (value 0.001)
T total kinetic energy (J)
T
e
elastic kinetic energy (J)
T
r
kinetic energy for rigid links (J)
t
d
delay time (s)
t time (s)
U total potential energy (J)
U
e
elastic potential energy for flexible links (J)
U
r
potential energy for rigid links (J)
u1x4 t 3 flexural displacement function
x state variables vector
2 x state derivatives vector
x
c
commanded x-endpoint position (m)
x actual x-endpoint position (m)
y
c
commanded y-endpoint position (m)
y actual y-endpoint position (m)
z measurement output [z
1
z
2
]
T
5
i j
elastic deformation (m)
2004 SAGE Publications. All rights reserved. Not for commercial use or unauthorized distribution.
at PENNSYLVANIA STATE UNIV on February 5, 2008 http://jvc.sagepub.com Downloaded from
CONTROL OF A TWO-LINK ROBOT MANIPULATOR 1417
damping ratio (value 0.707)
0 slew angle (rad)

0 joint rate (rad s


1
)
1 slew angle vector
joint rate vector
0
t
link distal end angle of rotation (rad)
. fuzzy logic system output parameter
.
ppi
L pinnedpinned ith mode characteristic root ( i v)
.
ci
L cantilever ith mode characteristic root (i 0.5)v
mass per unit length (kg m
1
)
, mass per unit volume (kg m
3
)
torque vector control law [
1

2
]
T

1
torque actuating link 1 (N-m)

2
torque actuating link 2 (N-m)

i
mode shape for ith mode
o
ci
cantilever elastic frequency mode (Hz)
o
ppi
pinnedpinned elastic frequency mode (Hz)
1. INTRODUCTION
Strict requirements exist for minimal vibration and precise control of long flexible robots
deployed in spacecraft operations. To study the effects of direction change disturbances and
methods of controlling the resulting flexural vibrations, a square trajectory provides an ideal
case for intense vibration effects requiring greater control effort at the four abrupt orthogonal
direction switches. For this purpose a 12.6 12.6 m
2
trajectory was chosen as suitable for
demonstrating tracking control.
The same square trajectory was used by Banerjee and Singhose (1998), who obtained
excellent results using an input shaping method coupled with an inverse kinematics control
scheme for both linear and nonlinear control laws, to simulate tracking control of a two-
link robot manipulator and reduce residual vibrations. Their full-order flexible dynamics
equations were derived using a recursive order-n algorithm in which each robot link was dis-
cretized into three rigid segments connected by torsion springs that are suitable for modeling
large bending deformations. Their work prompted the investigations presented in this paper.
Also, justification for the work presented in this paper lies in the desire to provide tracking
control accuracy for a model that closely represents actual flexible robot manipulator dy-
namics without resort to discretization or linearization typically used in finite-element and
state-space formulations.
Beres and Sasiadek (1995) formulated a Lagrange finite-element dynamic model of an
n-link flexible manipulator, in which large rigid-body rotational motion of the links and
their small elastic deformations are coupled. The dynamic model, derived for a two-link
flexible manipulator with two finite elements per link, may be extended to an arbitrary
n-link manipulator with n elements per link, as their simulation results show, but at greater
computational burden.
Sasiadek and Srinivasan (1989) applied a model reference adaptive control (MRAC)
technique to investigate position and vibration control of a distributed mass single-link flex-
2004 SAGE Publications. All rights reserved. Not for commercial use or unauthorized distribution.
at PENNSYLVANIA STATE UNIV on February 5, 2008 http://jvc.sagepub.com Downloaded from
1418 A. GREEN and J. Z. SASIADEK
ible manipulator using a modal expansion method based on a pinnedfree beamconfiguration
to determine the first three significant vibration modes. Their MRAC strategy, based on
state-space formulation, was an attempt to achieve smaller position errors and an alternative
to regular PID control that failed to give satisfactory results. They coupled an elastic poten-
tial function, derived from the modal expansion for a single flexible link with pinnedfree
boundary conditions, to the rigid-link dynamics and formulated a EulerLagrange dynam-
ics model and generalized inertia matrix of the manipulator. A simple feedback control law
comprises a derivative gain and velocity feedback term and a gravity compensation term.
PID transient responses to step and impulse inputs are compared to MRAC responses. Their
results demonstrate accuracy of the modal expansion method within an MRAC strategy to
achieve further reduction of positioning errors and decrease settling time of transient re-
sponse to step inputs. Their work, together with other references on the use of assumed
modes in formulating flexible dynamics models, provides support to the work presented in
this paper (Fraser and Daniel, 19911 Junkins and Kim, 19931 Mordfin and Tadikonda, 2000).
Lee et al. (1994) have applied a fuzzy logic approach to endpoint position control of an
experimental flexible single-link manipulator, where the fuzzy rules have position errors and
rates as inputs, with motor hub speed as the output variable to drive the physical manipulator.
Green and Sasiadek (2000, 2001, 2002) have studied tracking of a square trajectory
by a two-link robot manipulator using inverse dynamics, optimal and fuzzy logic system
control schemes based on rigid-link dynamics and a control law derived from proportional
and derivative gains computed using the dominant vibration mode. In particular, Green
and Sasiadek (2001) have provided results of simulations using a linear quadratic Gaussian
control scheme with an extended Kalman filter (EKF) and fuzzy logic adaptive extended
Kalman filter (FLAEKF) to control the endpoint tracking of a two-link robot manipulator.
Process and measurement noise are input to simulate inherent sensor, mechanical component
and dynamics noise with the EKF and FLAEKF each performing disturbance rejection to
achieve accurate tracking with various results. It was found that while the EKF tended to
provide divergent tracking at various noise intensities, the FLAEKF corrected the divergence
and consistently produced greater accuracy, although at a high computational time burden.
De Silva (1995) has used a fuzzy control technique to simulate tracking control of a
square trajectory by a rigid two-link robot manipulator where the fuzzy inference system is
used to tune the servo gains achieving precise control at direction switches.
Passino and Yurkovich (1990) have used joint angles and rates as fuzzy controller inputs
with voltage outputs applied to the actuators at each joint of an experimental two-link flexible
robot. Responses were obtained to commanded 90

slew angles for two control systems: one


with coupled fuzzy controllers and the other uncoupled.
Literature on single-link and multilink robot dynamics and control is extensive, of which
those referenced provide some insight to the difficulties encountered and techniques used to
control the endpoint of a robot manipulator and form a basis upon which further techniques
can be investigated.
The aim of this paper is to demonstrate control of a two-link robot manipulator modeled
with rigid and flexible dynamics based on dominant assumed modes for EulerBernoulli
pinnedpinned and cantilever beam boundary conditions. Of the three control schemes in-
vestigated with rigid-link dynamics, i.e. inverse dynamics, linear quadratic regulator (LQR)
and fuzzy logic, the inverse dynamics model is pursued further to model flexible-link ro-
2004 SAGE Publications. All rights reserved. Not for commercial use or unauthorized distribution.
at PENNSYLVANIA STATE UNIV on February 5, 2008 http://jvc.sagepub.com Downloaded from
CONTROL OF A TWO-LINK ROBOT MANIPULATOR 1419
Figure 1. Two-link robot manipulator.
bot dynamics and to devise a control technique to reduce or eliminate transient vibrations at
direction switches and achieve zero error tracking. Investigations on LQR and fuzzy logic
schemes using flexible-link dynamics are not considered further in this study.
2. ROBOT DYNAMICS
2.1. Robot Conf iguration
The two-link robot shown in Figure 1 has a shoulder joint revolute 21 rad and an elbow
joint oscillating 1 3221 rad. Robot motion and vibration modes are planar with gravity,
joint inertia, friction, and payload inertias being neglected. Robot parameters and physical
constants for the robot manipulator were taken from Banerjee and Singhose (1998).
2.2. Rigid-Link Robot Dynamics
A conventional closed form of the nonlinear dynamics of a two-link robot manipulator with
rigid links may be derived in terms of kinetic and potential energies stored in the system
by the EulerLagrange formulation (Asada and Slotine, 1986). Given an independent set of
generalized coordinates, q
i
2 q
1
3 q
2
3 4 4 4 q
n
, the total kinetic and potential energies, T and U
respectively, stored in the system may be defined by the Lagrangian
15q
i
3 3 q
i
6 2 T 4U i 2 13 4 4 4 3 n (1)
and the dynamic equations of motion for the system are derived in the form given by
d
dt
71
7 3 q
i
4
71
7q
i
2 F
i
3 i 2 13 4 4 4 3 n4 (2)
2004 SAGE Publications. All rights reserved. Not for commercial use or unauthorized distribution.
at PENNSYLVANIA STATE UNIV on February 5, 2008 http://jvc.sagepub.com Downloaded from
1420 A. GREEN and J. Z. SASIADEK
For the complete robot ensemble, the kinetic and potential energies for rigid links are
derived respectively as
T
r
=
1
2
n

k=1
r

p=1
M
kp
q
i
q
j
, i = 1, 2; j = 1, 2; n = 2; r = 2 (3)
and
U
r
=
n

i =1
m
i
g
T
r
0,C
i
, n = 2 (4)
where M
kp
is a kp element of the robot inertia matrix M. Substituting equations (3) and (4)
into equation (2) and applying the differential operators, we derive the EulerLagrange form
of the dynamic equations for rigid links as
= M(0)

0G
_

0, 0
_
. (5)
For equation (5), gravity terms are omitted for space borne operations, 2 substitutes for
F
i
, i.e. the actuating torque vector acting on the robot joints, M is the robot inertia matrix
and G is a coupling matrix comprising centrifugal and Coriolis force components given by
M =
_

_
m
1
L
2
3
m
2
L
2
_
4
3
cos 0
2
_
m
2
L
2
_
1
3

1
2
cos 0
2
_
m
2
L
2
_
1
3

1
2
cos 0
2
_
m
2
L
2
3
_

_
(6)
G =
_
1
2
m
2
L
2
sin 0
2
__

0
2
_
2

0
1


0
2
_

0
1
_
. (7)
2.3. Flexible-Link Robot Dynamics
In addition to the nonlinearity of rigid-link dynamics, achieving accurate tracking control
of the two-link robot with flexible links is compounded by deformation of the links, asso-
ciated flexural vibrations and non-minimum phase response. To enable design of a suitable
control strategy, the robot model must capture the nonlinear flexible dynamics of the ro-
bot and the control method must sufficiently dampen residual vibrations while eliminating
non-minimum phase response. Using assumed modes of vibration to model robot dynam-
ics captures the interaction between flexural vibrations and nonlinear multibody dynamics.
By assigning assumed modes for an EulerBernoulli beam with either cantilever or pinned
pinned boundary conditions to a two-link manipulator and by adding a flexibility potential
function in the EulerLagrange equations, a complete flexible dynamics model is derived
that describes a flexible manipulator suitable for simulation.
2004 SAGE Publications. All rights reserved. Not for commercial use or unauthorized distribution.
at PENNSYLVANIA STATE UNIV on February 5, 2008 http://jvc.sagepub.com Downloaded from
CONTROL OF A TWO-LINK ROBOT MANIPULATOR 1421
2.3.1. Assumed Modes Method
In the assumed modes method (Thomson, 19811 Fraser and Daniel, 19911 Junkins and
Kim, 19931 Robinett et al., 2002), the dynamic model of the robot manipulator is described
by a set of vibration modes other than its natural modes. Modeling flexural vibrations
of mechanical elements using assumed modes is a well-established technique. Using as-
sumed modes to model flexibility requires EulerBernoulli beam theory boundary condi-
tions and accommodates changes in configuration during operation, whereas natural modes
must be continually recomputed. Elastic deformations are modeled by a finite series of
space-dependent admissible functions multiplied by a specific set of time-dependent am-
plitude functions resulting in amplitudes that form the generalized configuration coordinates
in the EulerLagrange dynamics formulation. Admissible functions satisfy, at least, the sys-
tem geometric boundary conditions. A chosen set of admissible functions forms the basis
functions in the assumed modes method and are applied to the manipulator throughout its
operational workspace, provided geometric boundary conditions are consistent. An approxi-
mate deformation of any continuous elastic beam subjected to transverse vibrations is given
by
u(x, t ) =
n

i =1

i
(x)q
i
(t ) (8)
where
i
(x)denotes the assumed mode shape for specific beam boundary conditions. The
shape functions, u(x, t ), substitute into the EulerLagrange dynamics formulation given by
equations (1) and (2).
2.3.2. Mode Summation Method
For the general case of an EulerBernoulli beam subjected to a uniform distributed load
along its length L, the equation of motion is given by (Thomson, 1981)
c
2
cx
2
_
EI
c
2
u(x, t )
cx
2
_
dx m(x)
c
2
u(x, t )
ct
2
= p(x, t ), vt, 0xL (9)
for which the normal modes
i
must satisfy the following equation and its boundary condi-
tions
_
EI
//
i
_
//
o
2
i
m(x)
i
= 0 (10)
and the solution is given in terms of
i
(x) by equation (8) (Thomson, 1981).
Substituting equation (8) into equations (1) and (2), the generalized coordinates for beam
deformation can be determined and the elastic kinetic and potential energies are given by
T
e
=
1
2
n

i
n

j
q
i
q
j
_
L
0

j
m(x)dx =
1
2
n

i
M
i
q
2
i
i = 1, . . . , n, j = 1, . . . , n (11)
where M
i
is the generalized mass matrix dependent upon mass distribution and assumed
modes shapes defined as
2004 SAGE Publications. All rights reserved. Not for commercial use or unauthorized distribution.
at PENNSYLVANIA STATE UNIV on February 5, 2008 http://jvc.sagepub.com Downloaded from
1422 A. GREEN and J. Z. SASIADEK
M
i
=
_
L
0

2
i
(x)m(x)dx i = 1, . . . , n (12)
and
U
e
=
1
2
n

i
n

j
q
i
q
j
_
L
0
EI
//
i

//
j
dx =
1
2
n

i
K
i
q
2
i
=
1
2
n

i
o
2
i
M
i
q
2
i
i = 1, . . . , n, j = 1, . . . , n. (13)
Written in matrix form the equations for flexural motion of a non-actuated link are given
as
M q Kq = 0. (14)
Combining the flexural equations of motion for two links with those previously de-
rived for rigid links, we obtain the following flexible dynamics matrix equations (Fraser
and Daniel, 19911 Junkins and Kim, 19931 Robinett et al., 2002)
= M(q) q G( q, q) Kq. (15)
M is an inertia matrix of rigid and flexible-link elements, G is a dynamic coupling ma-
trix for both rigid and elastic Coriolis and centrifugal effects, and K is a diagonal matrix
of stiffness elements pertaining to flexural deflections of the links, presented in detail in
Fraser and Daniel (1991). The vector q is a set of generalized coordinates comprising joint
angle and flexible-link displacements. In applying the EulerBernoulli beam theory for cal-
culating assumed modes of vibration, small elastic deflections are an underlying assumption
where second-order terms of interacting elastic modes can be neglected. This, together with
the orthogonal properties of assumed modes, enables simplification of the dynamics matrix
equation (15). Omitting elastic Coriolis and centrifugal components gives the rigid dynamics
matrix G in equation (5).
2.4. Cantilever Assumed Modes
From transverse beam vibration theory (Timoshenko et al., 1974), cantilever mode shapes
are given by

ci
(x) = cosh .
ci
x cos .
ci
x k
ci
(sinh .
ci
x sin .
ci
x) (16)
where k
ci
=
cos .
ci
L cosh .
ci
L
sin .
ci
L sinh .
ci
L
(17)
for which .
ci
L - (i
1
2
)v, i = 1, . . . , n, are numerically approximated roots of the char-
acteristic equation
cos .
ci
L cosh .
ci
L 1 = 0. (18)
2004 SAGE Publications. All rights reserved. Not for commercial use or unauthorized distribution.
at PENNSYLVANIA STATE UNIV on February 5, 2008 http://jvc.sagepub.com Downloaded from
CONTROL OF A TWO-LINK ROBOT MANIPULATOR 1423
Modal frequencies are given by
o
ci
= (.
ci
L)
2
_
EI
L
4
. (19)
Proportional and derivative gains for the dominant assumed mode frequency are K
p
=
diag

o
2
c1
o
2
c1

=diag [150.79, 150.79] and K


d
=diag[2o
c1
2 o
c1
] =diag[17.364, 17.364]
where o
c1
= 12.28 Hz.
2.5. PinnedPinned Assumed Modes
Similarly, from transverse beam vibration theory pinnedpinned mode shapes are given by

ppi
=
_
2
L
sin .
ppi
x (20)
for which .
ppi
L = i v are roots of the characteristic equation sin .
ppi
L = 0, i = 1, . . . , n.
Modal frequencies are given by
o
ppi
= (.
ppi
L)
2
_
EI
L
4
. (21)
For pinnedpinned assumed modes, the distal end of a link coincides with that of a rigid
link with angle 0
t
subtended at the joint by the distal end. Proportional and derivative gains
for the dominant assumed mode frequency are given by K
p
= diag

o
2
pp1
o
2
pp1

= diag
[1188.4 1188.4] and K
d
= diag

2o
pp1
2 o
pp1

= dia[48.746 48.746] where o


pp1
=
34.474 Hz.
2.6. Choice of Assumed Modes
While many researchers generally prefer treating the links as cantilevered beams, there is
no stipulation as to which set of assumed modes should be used. Ideally, the optimum set
of assumed modes is that closest to natural modes of the system. Natural modes depend
on several factors within the robot system ensemble including size of hub inertia and size
of payload mass. For large joint gearing inertia and relatively small payload mass, the link
may be considered clamped at the joint with the endpoint free. Conversely, for smaller joint
gearing inertia and larger payload mass both ends of the link may be considered pinned. The
ultimate choice requires an assessment based on the actual robot structure and anticipated
range of payloads together with its natural modes (Fraser and Daniel, 1991).
The use of both cantilever and pinnedpinned modes in this paper demonstrates the
behavior of each for the robot structure aforementioned, although only link mass without
hub inertia and payload has been included in the robot dynamics equation (15).
2004 SAGE Publications. All rights reserved. Not for commercial use or unauthorized distribution.
at PENNSYLVANIA STATE UNIV on February 5, 2008 http://jvc.sagepub.com Downloaded from
1424 A. GREEN and J. Z. SASIADEK
Figure 2. Inverse 1exible dynamics control scheme.
2.7. Non-Minimum Phase Response
An inherent problem met in the control of slender flexible-link robots is non-minimum phase
response (Alexander, 1988). A characteristic of this phenomenon exists when joint actua-
tion response delays occur in reaction to a control input opposite in sense to that expected.
Practically, when torque actuates a joint it induces flexing and momentary acceleration of
the endpoint in a direction opposite in sense to that commanded by the actuating torque. In
analytical control theory, this behavior is described by a pole or zero of the transfer func-
tion occurring in the right-half s-plane and termed a phase shift, or transport lag, between
the actuator and link endpoint. For continuous systems, this phase lag creates a time delay
between the actual endpoint position and control actuation of the joint to correct feedback
position errors and corresponds to the time taken by mechanical waves to propagate through
the link from joint to endpoint.
Closely associated with non-minimumphase response is the distance between sensor and
actuator, or non-collocation. This causes time delays in joint control actuation in response
to feedback position errors based on position data from a sensor located at the endpoint.
Collocated control provides joint actuation in response to joint angle data from a sensor
located at the joint. Typically, collocated control is suitable for fixed-base industrial robot
manipulators with rigid links and controlled at speeds where flexibility may be ignored.
Also, non-collocation for these manipulators does not encounter flexibility because the link
is assumed rigid. However, for slender flexible robot manipulators, there is significant link
deformation at the endpoint necessitating a technique to provide accurate control by adjusting
for the phase lag between actuator and endpoint.
An operational-space control scheme provides a suitable model for demonstrating non-
collocated control as the joint rates and angles are transformed through direct kinematics
equations into Cartesian endpoint positions and velocities for generating feedback errors for
the control law computation.
The control schemes presented in this paper are all of the operational-space type. Non-
minimum phase response is modeled by implementing a time delay in the feedback loop
of the inverse flexible dynamics control schemes of Figures 2 and 3 (Dutton et al., 1997).
Actual transport delay blocks (not shown) with a second-order Pad approximation were
implemented in the Matlab/Simulink
TM
models. To correct for this non-minimum phase
response, the same time delay acts upon the control law input to the robot dynamics to
2004 SAGE Publications. All rights reserved. Not for commercial use or unauthorized distribution.
at PENNSYLVANIA STATE UNIV on February 5, 2008 http://jvc.sagepub.com Downloaded from
CONTROL OF A TWO-LINK ROBOT MANIPULATOR 1425
Figure 3. Inverse 1exible dynamics fuzzy adapted control scheme.
achieve minimum phase response. The time delay was determined using the transverse beam
vibration wave velocity equation given by
Wave velocity c =
_
E
,
=
_
1745833
21
= 288.33 ms
1
. (22)
Delay time from joint 1 to endpoint for two link lengths, i.e. 9 m, is given by
t
d1
=
9
288.33
= 0.0312 s. (23)
Delay time from joint 2 to endpoint for one link length, i.e. 4.5 m, is given by
t
d2
=
4.5
288.33
= 0.0156 s. (24)
The average trajectory simulation time is 402 s for 16,000 steps at 0.001 step size, i.e.
0.0252 s per step:
1 Simulation delay time for joint 1
d
1
=
0.0312
0.0252
= 1.238 ss; (25)
1 Simulation delay time for joint 2
d
2
=
0.0156
0.0252
= 0.619 ss. (26)
As the joint rates and angles are transformed to endpoint positions immediately after
output from the robot manipulator dynamics, and for simulation purposes, the time delay is
implemented in the position feedback loop at the maximum delay time d
1
, i.e. worst case.
Time delays are not considered for the inverse rigid dynamics, LQR and fuzzy control
schemes because their robot dynamics models are based on rigid links.
2004 SAGE Publications. All rights reserved. Not for commercial use or unauthorized distribution.
at PENNSYLVANIA STATE UNIV on February 5, 2008 http://jvc.sagepub.com Downloaded from
1426 A. GREEN and J. Z. SASIADEK
3. CONTROL SCHEMES
Three control schemes considered in this paper are based on inverse dynamics, state-space
and fuzzy logic models of the rigid-link robot dynamics while two additional control schemes
extend the inverse dynamics scheme to include flexible-link robot dynamics. The distinction
between the two inverse flexible dynamics schemes is that one has a fuzzy logic system
implemented in its feedback loop for link deformations and provides a scalar output variable
that adapts the control law in response to the magnitude of deformation of each link.
3.1. Control Law
An intuitive Jacobian transpose control law applied to all control schemes provides a joint
actuating torque vector given by
2 = J
T
(0)
_
K
p
_
e
x
e
y
__
K
d
_
e
x
e
y
_
. (27)
Proportional and derivative gain values K
p
and K
d
, calculated using cantilever and
pinnedpinned assumed mode frequencies, are used in the following control schemes. Com-
mon to all control schemes is the use of output slew angles 0
1
, 0
2
transformed in the direct
kinematics equations (28) and (29) to x, y endpoint positions and fed back to form position
errors x
c
x, y
c
y:
x = L
1
cos(0
1
) L
2
cos(0
1
0
2
) (28)
y = L
1
si n(0
1
) L
2
si n(0
1
0
2
). (29)
The resulting errors are fed forward for computing the control laws of equation (27) and
(35).
3.2. Inverse Dynamics Scheme
The inverse rigid dynamics control scheme shown in Figure 4 was investigated and presented
in previous work by Green and Sasiadek (2000, 2002). Using the control law of equation
(27), the torque vector feeds into the nonlinear inverse dynamics equations giving an
angular acceleration vector output then double integrated to obtain joint rates

0
1
,

0
2
and slew
angles 0
1
, 0
2
. Joint rates transform to endpoint velocities and feed back to form velocity
errors x
c
x, y
c
y. Joint rates and slew angles also feed back into the inverse dynamics
equations. The control scheme is typical for a two-link robot manipulator in the literature on
robot analysis and control (Asada and Slotine, 1986).
3.3. Linear Quadratic Regulator Scheme
The LQR control scheme shown in Figure 5 was investigated and presented in previous work
by Green and Sasiadek (2000, 2001, 2002). It has a similar feedback structure as the inverse
2004 SAGE Publications. All rights reserved. Not for commercial use or unauthorized distribution.
at PENNSYLVANIA STATE UNIV on February 5, 2008 http://jvc.sagepub.com Downloaded from
CONTROL OF A TWO-LINK ROBOT MANIPULATOR 1427
Figure 4. Inverse rigid dynamics control scheme.
Figure 5. LQR control scheme.
dynamics control scheme but the dynamics equations given by equation (5) are linearized
and transformed into the state-space form of equations (30) and (31). The torque vector
computed by the control law of equation (27) feeds through LQ input gains into the state-
space equations (30) and (31), to actuate each joint:
x = Ax Bu (30)
z = Cx Du. (31)
The state vector x has two joint angles and two angular velocities for rigid-link dynamics.
The state feedback system minimizes the performance index given by equation (32):
J =
o
_
0
_
x
T
Qx u
T
Ru
_
dt. (32)
The LQ gain matrix K is calculated by
K = R
1
B
T
P (33)
where P is a definite positive solution of the Riccati equation:
A
T
P PAPBR
1
B
T
Q = 0. (34)
2004 SAGE Publications. All rights reserved. Not for commercial use or unauthorized distribution.
at PENNSYLVANIA STATE UNIV on February 5, 2008 http://jvc.sagepub.com Downloaded from
1428 A. GREEN and J. Z. SASIADEK
Figure 6. Fuzzy logic control scheme: (a) control scheme2 (b) fuzzy robot manipulator.
Q = R = diag [1, 1, 1, 1] are arbitrary weighting matrices of the performance index.
Outputs from the state-space dynamics feed through LQ gains to provide slew angles
0
1
, 0
2
.
3.4. Fuzzy Logic Scheme
The fuzzy logic scheme shown in Figure 6 was investigated and presented in previous work
by Green and Sasiadek (2000, 2002). Two coupled fuzzy logic systems model the manipu-
lator links and substitute for the inverse dynamics equations. The torque vector computed
by the control law of equation (27) feeds forward to each fuzzy link through normalizing
gains, K
N1
and K
N2
. Link 1 has a single torque input, and link 2 has acceleration and torque
inputs as shown in Figure 6(b). Acceleration output from link 1 is input to link 2 through
normalizing gain K
N3
to create a coupling effect between themand capture the combined mo-
tion and interaction of both links as the acceleration of link 1 at the elbow joint significantly
affects link 2 endpoint vibration. Conversely, acceleration at the endpoint of link 2 does not
significantly affect the motion of link 1.
The fuzzy logic systems have input and output variables typically each with nine Gaussian
membership functions shown in Figures 7(a)(d) for cantilever links. Verbal descriptors used
in the fuzzy rules are composed using the notation P = positive, N = negative, V = very, H
= high and L = low to denote the membership functions NVH, NH, NL, NVL, ZERO, PVL,
PL, PH and PVH, respectively. For cantilever assumed modes, link 1 has nine fuzzy rules
with torque universe of discourse 500 to 500 N-m. Link 2 has 81 fuzzy rules with acceler-
ation and torque universes of discourse 2 to 2 rad s
2
and 200 to 200 N-m, respectively.
2004 SAGE Publications. All rights reserved. Not for commercial use or unauthorized distribution.
at PENNSYLVANIA STATE UNIV on February 5, 2008 http://jvc.sagepub.com Downloaded from
CONTROL OF A TWO-LINK ROBOT MANIPULATOR 1429
Figure 7. Fuzzy membership functions (cantilever): (a) fuzzy link 1 torque input2 (b) fuzzy link 2 torque
input2 (c) fuzzy link 2 acceleration input2 (d) fuzzy link 1 and 2 acceleration outputs2 (e) fuzzy logic
system 1exible-link deformation input variables o
1
and o
2
2 (f) fuzzy logic system output variable ..
Each link has acceleration output universe of discourse 5 to 5 rad s
2
. For pinnedpinned
assumed modes, the universes of discourse are respectively 5000 to 5000 N-m for torque
actuating link 1, 1000 to 1000 N-m for torque actuating link 2, 2 to 2 rad s
2
for coupling
acceleration input to link 2 and 5 to 5 rad s
2
for acceleration outputs from both links.
Table 1 is a matrix of fuzzy rules of the form:
IF torque 1 is NL THEN acceleration 1 is NL1
IF torque 1 is PH THEN acceleration 1 is PH.
Table 2 is a matrix of fuzzy rules of the form:
IF torque 2 is PL and acceleration 12 is PL THEN acceleration 2 is PH1
IF torque 2 is NVL and acceleration 12 is PVL THEN acceleration 2 is ZERO.
2004 SAGE Publications. All rights reserved. Not for commercial use or unauthorized distribution.
at PENNSYLVANIA STATE UNIV on February 5, 2008 http://jvc.sagepub.com Downloaded from
1430 A. GREEN and J. Z. SASIADEK
Table 1. Rule matrix for link 1.
Torque (1) NVH NH NL NVL ZERO PVL PL PH PVH
Acceleration (1) NVH NH NL NVL ZERO PVL PL PH PVH
Table 2. Rule matrix for link 2.
Output accelerations feed through output scaling gains, K
1
and K
2
, adjusted to modify
the base widths of membership functions and dampen robot link vibrations to obtain a square
trajectory with greatest precision. The fuzzy logic scheme for cantilever and pinnedpinned
mode frequencies produced initial and final square trajectories with the following stability
limits of output scaling gains of K
1
and K
2
:
for cantilever
8000 1 K
1
1 200000
2000 1 K
2
1 80000
for pinnedpinned
500000 1 K
1
1 10000001
50000 1 K
2
1 100000
Outputs from the fuzzy robot manipulator provide slew angles 2
1
3 2
2
.
3.5. Inverse Flexible Dynamics Control Scheme
The control scheme shown in Figure 2 comprises the same control architecture elements as
the inverse rigid dynamics control scheme shown in Figure 4, but comprises inverse flexible
dynamics equations with link deformations 4
1
3 4
2
fed back into the dynamics equations to-
gether with slew angles 2
1
,2
2
and joint rates
2
2
1
3
2
2
2
. Using the control law of equation (27),
the torque vector 5 feeds into the inverse flexible dynamics equations derived using an as-
sumed modes model (Mordfin and Tadikonda, 2000) to give a joint acceleration vector output
that is double integrated to obtain joint rates and slew angles. Joint rates
2
2
1
3
2
2
2
transform to
endpoint velocities and feedback to form velocity errors 2 x
c
3 2 x3 2 y
c
3 2 y. Errors feed forward
for computing the control law of equation (27). Joint rates and slew angles also feed back
into the inverse flexible dynamics equations.
2004 SAGE Publications. All rights reserved. Not for commercial use or unauthorized distribution.
at PENNSYLVANIA STATE UNIV on February 5, 2008 http://jvc.sagepub.com Downloaded from
CONTROL OF A TWO-LINK ROBOT MANIPULATOR 1431
Table 3. Fuzzy logic system rule matrix for adaptive control law.
o
2
NVH NH NL NVL ZERO PVL PL PH PVH
.
NVH PMAX PVVH PVVH PVVH PVH PVH PVH PH PM
NH PVVH PVVH PVH PVH PVH PH PH PM PH
NL PVVH PVH PVH PVH PH PH PM PM PH
NVL PVH PVH PH PH PH PM PM PM PL
o
1
ZERO PVH PH PL PVL ZERO PVL PL PH PVH
PVL PL PM PM PM PH PH PH PVH PVH
PL PM PM PM PH PH PVH PVH PVH PVVH
PH PM PM PH PM PVH PVH PVH PVVH PVVH
PVH PM PH PH PVH PVH PVVH PVVH PVVH PMAX
3.6. Fuzzy Adapted Inverse Flexible Dynamics Control Scheme
The control scheme shown in Figure 3 is similar to Figure 2 but modified to include a fuzzy
logic system. Adaptive control is achieved by the scalar variable . output from the fuzzy
logic system to adapt the Jacobian transpose control law. The variable . is determined by
elastic deformations o
1
and o
2
fed back from the control scheme outputs through a normal-
izing gain to input the fuzzy logic system. The resulting control law is
= .
_
J
T
(0)
_
K
p
_
e
x
e
y
_
K
d
_
e
x
e
y
___
. (35)
Fuzzy logic system input variables o
1
, o
2
and output variable . each have nine Gaussian
membership functions shown in Figures 7(e) and (f). Verbal descriptors used in the fuzzy
rules are composed using the notation P = positive, N = negative, V = very, MAX = max-
imum, H = high, and L = low to denote the membership functions NVH, NH, NL, NVL,
ZERO, PVVL, PVL, PL, PH, PVVH and PMAX, respectively. Universes of discourse range
from 5 to 5 m for input variables o
1
, o
2
and from 0 to 20 for output variable ..
Table 3 is a matrix of fuzzy rules of the form:
IF o
1
is NL and o
2
is PL THEN . is PM1
IF o
1
is PVL and o
2
is PH THEN . is PVH.
The fuzzy logic system is developed intuitively as the magnitude of elastic deformation
of either link varies positively or negatively, so it may complement or counter the deforma-
tion of the other link. The value of . adjusts according to the severity of resultant deforma-
tion.
The control law torque vector given by equation (35) feeds into the inverse flexible dy-
namics equations using an assumed modes model derived by Mordfin and Tadikonda (2000)
to give a joint acceleration vector output then double integrated to obtain link deformations
o
1
, o
2
, joint rates

0
1
,

0
2
and slew angles 0
1
, 0
2
. Joint rates transform to endpoint velocities
and feedback to form velocity errors x
c
x, y
c
y. Errors feed forward for computing
the control law of equation (35). Joint rates and slew angles also feedback into the inverse
dynamics equations.
2004 SAGE Publications. All rights reserved. Not for commercial use or unauthorized distribution.
at PENNSYLVANIA STATE UNIV on February 5, 2008 http://jvc.sagepub.com Downloaded from
1432 A. GREEN and J. Z. SASIADEK
Figure 8. Rigid-link dynamics trajectories: (a) cantilever inverse dynamics model2 (b) pinnedpinned
inverse dynamics model2 (c) cantilever LQR state-space model2 (d) pinnedpinned LQR state-space
model2 (e) cantilever fuzzy logic model2 (f) pinnedpinned fuzzy logic model.
4. RESULTS
Tracking starts at the bottom-left corner of each trajectory with link 1 rotating 2v rad clock-
wise for which results are shown in Figures 8, 9, 10, and 11 Trajectories obtained for the
inverse rigid dynamics control scheme using cantilever and pinnedpinned assumed mode
PD gains are shown in Figures 8(a) and (b). It is evident that large amplitude transients
are experienced at the direction switches for the dominant cantilever assumed mode. In
contrast, greater control accuracy with lower amplitude transients is obtained using the dom-
inant pinnedpinned assumed mode. Tracking times are 220 and 212 s for cantilever and
pinnedpinned modes, respectively. Trajectories obtained for the LQR scheme, with a rigid
dynamics model using cantilever and pinnedpinned assumed modes, are shown in Figures
8(c) and (d).
2004 SAGE Publications. All rights reserved. Not for commercial use or unauthorized distribution.
at PENNSYLVANIA STATE UNIV on February 5, 2008 http://jvc.sagepub.com Downloaded from
CONTROL OF A TWO-LINK ROBOT MANIPULATOR 1433
Figure 9. Rigid-link dynamics trajectory comparisons at the 3rst (top left) direction switch: (a) cantilever2
(b) pinnedpinned.
The LQR scheme produces a much-improved result over the inverse dynamics scheme
for both cantilever and pinnedpinned assumed modes, although the trajectory is better for
the pinnedpinned assumed mode. Tracking times are 188 and 372 s for cantilever and
pinnedpinned modes, respectively.
Results for the fuzzy logic scheme shown in Figures 8(e) and (f) demonstrate that greater
control accuracy can be achieved by substituting the dynamics equations with a fuzzy logic
system representing each link and the benefit of flexible tuning of fuzzy output scaling gains
K
1
and K
2
, within their limits of stability to dampen the vibration transients, particularly
for the cantilever mode. Tracking times are 352 and 360 s for cantilever and pinnedpinned
modes, respectively.
A comparison of transients zoomed at the first direction switch for inverse dynamics,
LQR and fuzzy control schemes using rigid dynamics models are shown in Figures 9(a) and
(b). For the cantilever mode, the inverse dynamics control scheme maximum overshoot is
2004 SAGE Publications. All rights reserved. Not for commercial use or unauthorized distribution.
at PENNSYLVANIA STATE UNIV on February 5, 2008 http://jvc.sagepub.com Downloaded from
1434 A. GREEN and J. Z. SASIADEK
Figure 10. Inverse 1exible dynamics trajectories: (a) cantilever2 (b) pinnedpinned2 (c) cantilever
trajectory comparison at the 3rst (top left) direction switch2 (d) pinnedpinned trajectory comparison at
3rst (top left) direction switch2 (e) cantilever trajectory comparison (zoomed)2 (f) pinnedpinned trajectory
comparison (zoomed).
0.7 m at 4600 ss and settling time of 6000 ss. The maximum overshoot for the LQR control
scheme is much lower at 0.3 m at 4400 ss and settling time of 5800 ss. The most effective
control is obtained with the fuzzy logic scheme for which the result has a lower maximum
overshoot of 0.2 m at 4250 ss immediately followed by a smooth gradual decay in a settling
time of 5800 ss.
In comparison, for the pinnedpinned mode, the inverse dynamics scheme exhibits a
maximum overshoot of 0.1 m at 4180 ss with a settling time of 4800 ss. For the LQR control
scheme, the maximum overshoot is a much lower 0.1 m at 4100 ss with a settling time of
4400 ss. For the fuzzy logic scheme, the peak amplitude occurs below the commanded 6.3 m
position at 6.1 m and 4100 ss, i.e. 0.2 m before reaching the commanded position of 6.3 m,
and rises gradually to achieve it in a settling time of 1000 ss.
2004 SAGE Publications. All rights reserved. Not for commercial use or unauthorized distribution.
at PENNSYLVANIA STATE UNIV on February 5, 2008 http://jvc.sagepub.com Downloaded from
CONTROL OF A TWO-LINK ROBOT MANIPULATOR 1435
Figure 11. Fuzzy adapted inverse 1exible dynamics trajectories: (a) cantilever2 (b) pinnedpinned2 (c)
cantilever trajectory comparison at the 3rst (top left) direction switch2 (d) pinnedpinned trajectory com-
parison at 3rst (top left) direction switch2 (e) cantilever trajectory comparison (zoomed)2 (f) pinnedpinned
trajectory comparison (zoomed).
For the pinnedpinned mode, the LQR scheme provides greater control accuracy in
achieving the desired trajectory with greater damping of vibration amplitude in the short-
est time. However, the fuzzy logic scheme produces the best result for the cantilever mode
but it provides overdamping for the pinnedpinned mode and fails to reach its commanded
position prior to convergence along the trajectory after the direction switch.
Endpoint velocities in both x and y directions are presented in Figures 12(a)(d) to com-
pare the changes in velocity along the trajectory between the cantilever and pinnedpinned
modes for the three rigid dynamics control schemes.
2004 SAGE Publications. All rights reserved. Not for commercial use or unauthorized distribution.
at PENNSYLVANIA STATE UNIV on February 5, 2008 http://jvc.sagepub.com Downloaded from
1436 A. GREEN and J. Z. SASIADEK
Figure 12. Endpoint velocities: (a) cantilever x-velocity1 (b) pinnedpinned x-velocity1 (c) cantilever
y-velocity1(d) pinnedpinned y-velocity.
The endpoint velocity for both sets of assumed modes exhibits sharp increases as it
approaches each direction switch for all control schemes, with the inverse dynamics scheme
being the most pronounced. The pinnedpinned mode exhibits a faster decrease in velocity
after each direction switch compared to the slower decrease of the cantilever mode. Peak
velocities are also higher for the pinnedpinned mode.
To exemplify the robot joint dynamics, joint accelerations are shown only for the in-
verse dynamics control scheme in Figures 13(a)(d) for both sets of assumed modes. The
cantilever mode accelerations are much higher and take longer to diminish than those of the
pinnedpinned mode.
Results for the inverse flexible dynamics model control scheme are shown in Figures
10(a)(f) and include trajectories tracked to compare simulations with and without time de-
lays to represent minimum phase, non-minimum phase, and corrected responses. While
the trajectories in Figures 10(a)(d) show consistent results with and without time delays,
Figures 10(e) and (f) more clearly show the results of implementing time delays in the feed-
back loop to affect non-minimum phase response and in the control input to affect corrective
control action. As a result of feedback time delay, it is evident that the transient response
overshort increases by 0.05 m and is minimally corrected 0.001 m by the control action
time delay. The cantilever mode trajectories show markedly higher maximum overshoots
2004 SAGE Publications. All rights reserved. Not for commercial use or unauthorized distribution.
at PENNSYLVANIA STATE UNIV on February 5, 2008 http://jvc.sagepub.com Downloaded from
CONTROL OF A TWO-LINK ROBOT MANIPULATOR 1437
Figure 13. Inverse rigid dynamics joint accelerations: (a) cantilever joint 11(b) pinnedpinned joint 11(c)
cantilever joint 21(d) pinnedpinned joint 2.
of 0.5 m and more severe oscillations compared to the 0.1 m maximum overshoots for the
pinnedpinned mode. Figures 14(a)(d) and Figures 15(a)(d) showno noticeable difference
in the endpoint velocities and joint accelerations. It is evident that assumed mode flexible
dynamics cause greater oscillation in the control action to counter the effects of flexibility
in the links. Flexural vibrations dominate variations in control action and nullify the ef-
fect of non-minimum phase response. Tracking times are 400 and 404 s for cantilever and
pinnedpinned modes, respectively.
The results obtained using the fuzzy adapted inverse flexible dynamics control scheme
demonstrate a significant reduction in vibration for both cantilever and pinnedpinned modes,
as shown in Figures 11(a)(d). The effects of implementing time delays to simulate non-
minimum phase response and its correction, shown in Figures 11(e) and (f), are similar to
results obtained for the prior case and considered insignificant in contrast to the overriding
effect of fuzzy adapted control action. The transient response for the pinnedpinned mode
exhibits a maximum overshoot of 0.05 m and a rapid decay to a steady state. In contrast,
the cantilever mode exhibits a maximum overshoot of 0.1 m and vigorous oscillations of
+/ 0.1 m for about one-third of each leg of the trajectory. Endpoint velocities are the same
for both cantilever and pinnedpinned modes with greatly reduced oscillation peaks. Simi-
2004 SAGE Publications. All rights reserved. Not for commercial use or unauthorized distribution.
at PENNSYLVANIA STATE UNIV on February 5, 2008 http://jvc.sagepub.com Downloaded from
1438 A. GREEN and J. Z. SASIADEK
Figure 14. Inverse 1exible dynamics endpoint velocities: (a) cantilever x-velocity2 (b) pinnedpinned
x-velocity2 (c) cantilever y-velocity2 (d) pinnedpinned y-velocity.
larly, joint accelerations for both assumed modes show no difference and demonstrate small
variation in control action, except at direction switches where pronounced yet instantaneous
peaks occur. Tracking times are 1064 and 1060 s for cantilever and pinnedpinned modes,
respectively. Matlab/Simulink
TM
, Control Systems and Fuzzy Logic Tollboxes were used
for all simulations.
5. CONCLUSIONS
In this paper we have demonstrated the relative effectiveness in tracking control of a two-
link robot manipulator using a rigid dynamics model and PD gains computed on dominant
cantilever and pinnedpinned assumed mode frequencies by three different techniques: in-
verse dynamics, LQR, and fuzzy logic. Of these, fuzzy logic produces the best result for the
cantilever mode and LQR for the pinnedpinned mode. The cantilever mode presents a more
severe case for control and the fuzzy logic scheme provides effective and accurate control
over damps for the pinned-pinned mode case.
The rigid dynamics model was used initially to establish a benchmark in systemresponse
and to ascertain the most suitable control scheme for further study of a flexible dynamics
model. The inverse dynamics scheme was chosen as it more closely represents the nonlinear
2004 SAGE Publications. All rights reserved. Not for commercial use or unauthorized distribution.
at PENNSYLVANIA STATE UNIV on February 5, 2008 http://jvc.sagepub.com Downloaded from
CONTROL OF A TWO-LINK ROBOT MANIPULATOR 1439
Figure 15. Inverse 1exible dynamics joint accelerations: (a) cantilever joint 12 (b) pinnedpinned joint 12
(c) cantilever joint 22 (d) pinnedpinned joint 2.
dynamical behavior of a flexible manipulator and presents a formidable challenge for con-
trol. Alternatively, the LQR scheme, based on a linear state-space dynamic model and the
fuzzy logic model, requires further development to capture multibody nonlinearity and link
flexibility.
In principle, results obtained for the inverse dynamics model for flexible links endorse
the theory of coupling large rotational rigid-link motions with small flexural deformations
to characterize the behavior of a two-link flexible robot, albeit restricted to the dominant
mode. Modeling non-minimum phase response by a time delay further characterizes the
complexity of the manipulators dynamical behavior and presents an additional challenge for
control. This is shown to be effectively remedied by time-delayed control action.
Implementing a time delay in the control action achieves minimal correction to the non-
minimum phase response. However, for the overriding problem of residual vibrations oc-
curring at direction switches, significant results are achieved by implementing a fuzzy logic
system that adapts the control action in response to the magnitude of link deformation.
The investigations culminate with a simple, yet effective, technique utilizing a hybrid
of inverse dynamics and fuzzy logic to achieve precision control of a flexible robot without
linearizing the flexible nonlinear dynamics equations.
Future work could pursue the LQR and fuzzy control schemes for flexible dynamics and
all three control schemes with flexible dynamics models for higher assumed modes, together
with the development of suitable control laws to achieve precision tracking control. Physical
2004 SAGE Publications. All rights reserved. Not for commercial use or unauthorized distribution.
at PENNSYLVANIA STATE UNIV on February 5, 2008 http://jvc.sagepub.com Downloaded from
1440 A. GREEN and J. Z. SASIADEK
laboratory flexible robot manipulator experiments could be performed with different joint
hub inertias and endpoint payloads to correlate with simulation results.
REFERENCES
Alexander, H. L., 1988, Control of articulated and deformable space structures, in Machine Intelligence and
Autonomy for Aerospace Systems, E. Heer and H. Lum, eds, Progress in Astronautics and Aeronautics,
AIAA, Washington, DC, pp. 327347.
Asada, H. and Slotine, J.-J. E., 1986, Robot Analysis and Control, Wiley, NY.
Banerjee, A. K. and Singhose, W., 1998, Command shaping in tracking control of a two-link flexible robot,
Journal of Guidance, Dynamics and Control 21 (6), 10121015.
Beres, W. and Sasiadek, J. Z., 1995, Finite element dynamic model of multilink flexible manipulators, Applica-
tions in Mathematics and Computer Science 5 (2), 231262.
de Silva, C. W., 1995, Intelligent Control: Fuzzy Logic Applications, CRC Press, Boca Raton, FL.
Dutton, K., Thompson, S., and Barraclough, B., 1997, The Art of Control Engineering, Addison-Wesley, Reading,
MA.
Fraser, A. R. and Daniel, R. W., 1991, Perturbation Techniques for Flexible Manipulators, Kluwer International
Series in Engineering and Computer Science, Vol. 138, Kluwer, Boston, MA.
Green, A. and Sasiadek, J. Z., 2000, Direct, optimal, stochastic and fuzzy control of a two-link flexiblemanipulator,
in Proceedings of the AIAA Guidance, Navigation and Control Conference and Exhibit, Denver, CO, August
1417, Paper No. AIAA 2000-4375.
Green, A. and Sasiadek, J. Z., 2001, Regular and extended Kalman filtering for a two-link flexible robot manip-
ulator, in Proceedings of the AIAA Guidance, Navigation and Control Conference and Exhibit, Montreal,
Quebec, Canada, August 69, Paper No. AIAA 2001-4565.
Green, A. and Sasiadek, J. Z., 2002, Methods of trajectory tracking for flexible-link manipulators, in Proceedings
of the AIAA Guidance, Navigation and Control Conference and Exhibit, Monterey,CA, August 58, Paper
No. AIAA 2002-4565.
Junkins, J. L. and Kim, Y., 1993, Introduction to Dynamics and Control of Flexible Structures, Education Series,
J. S. Przemieniecki, Editor-in-Chief, AIAA, Washington, DC.
Lee, J. X., Vukovich, G., and Sasiadek, J. Z.., 1994, Fuzzy control of a flexible-link manipulator, in Proceedings
of the American Control Conference, Baltimore, MD, June, Paper WM7-1:50.
Mordfin, T. G. and Tadikonda, S. S. K., 2000, Truth models for articulating flexible multibody dynamic systems,
AIAA Journal of Guidance, Dynamics and Control 23 (5), 805811.
Passino, K. M. and Yurkovich, S., 1990, Fuzzy Control, Addison-Wesley, Reading, MA.
Robinett, R. D. III, Dohrmann, C., Eisler, G. R., Feddema, J., Parker, G. G., Wilson, D. G., and Stokes, D., 2002,
Flexible Robot Dynamics and Controls, International Federation for Systems Research International Series
on Systems Science and Engineering, Vol. 19, Kluwer, Dordrecht.
Sasiadek, J. Z. and Srinivasan, R., 1989, Dynamic modeling and adaptive control of a single-link flexible manip-
ulator, AIAA Journal of Guidance, Dynamics and Control 12 (6), 838844.
Timoshenko, S., Young, D. H., and Weaver, W. Jr., 1974, Vibration Problems in Engineering, 4th edition, Wiley,
NY.
Thomson, W. T., 1981, Theory of Vibration with Applications, 2nd edition, Prentice-Hall, Englewood Cliffs, NJ.
2004 SAGE Publications. All rights reserved. Not for commercial use or unauthorized distribution.
at PENNSYLVANIA STATE UNIV on February 5, 2008 http://jvc.sagepub.com Downloaded from

Potrebbero piacerti anche