Beruflich Dokumente
Kultur Dokumente
TABLEIll
EXPERIMENTAL
VALUESFOR LOOPSOF DIFFERENT
SPEEDRATIOS
a = 0.1
Primaryloop
closed
0.30
1.97
tu2
ku2
primaryloop
open
0.30
1.97
= 0.8
lWP
closed
2.34
2.03
loop
open
2.42
1.97
and the ultimate parameters obtained are then used to auto-tune the
PUPID controller for the primary loop. The ratio of the ultimate
frequencies obtained in these two steps indicates the ratio of the
speeds of the loops and hence can be used to confirm the effectiveness
of cascade control. Subsequent re-tuning of either controller, if
required,can be done entirely in closed loop.
RWERENCES
G. Stephanopoulos, Chemical Process Conml: An Introduction to Theory and Practice. New Jersey: Prentice Hall, 1984.
P. W. Munill, Application Concepts Process Control. RTP, North
0.2
0.3
0.4
a5
0.6
0.7
0.8
0.9
Fig. 6. Plot of
against a.
Engineering Journal.
J. G. Ziegler and N. B. Nichols, Optimum settings for automatic
controllers, Trans. ASME, vol. 64, pp. 759-768, 1942..
C. C. Hang, K.J. Astrijm and W.K.Ho, Refinements of the ZieglerNichols tuning formula, IEE Proc. D, Control Theory and Application,
IV. CONCLUSION
A method is presented here to tune controllers in cascade loops.
It involves the use of on-off relays to obtain the ultimate gains and
frequencies of the loops when they undergo limit cycle oscillation.
The secondary loop is put on relay feedback first and a P or PI
controller is auto-tuned using established tuning rules. The primary
loop is then placed on relay feedback with the secondary loop closed
I.
INTRODUCTION
46
41
Fig. 3.
DOF.
B. The Cubic Conjguration
Although Stewart platforms have unique advantages over conventional active vibration control structures, the kinematics and dynamic
behavior of a Stewart platform is, in general, quite complex. The
primary difficulty with a general Stewart Platform is that motions
along different axes are strongly coupled and a motion in any
Cartesian direction requires motion of all of the legs, resulting in
mathematical complexity in control system design. As a result of
our study, we have discovered a cubic configuration of a Stewart
Platform which, for vibration control applications, eliminates much
of this problem.
There are an infinite number of possible geometric configurations
of a Stewart platform, depending upon the selection of the positions
of attachment points between actuators and plates, the physical size
of the mobile and base plate, as well as the range of actuator
displacement. Different configurations display different kinematics
relationships and dynamic behaviors. For example, in certain configurations, the coupling effects between individual actuators can
significantly degrade control performance, and it is even possible for
4s
.I
d L = J-ldX
(2.1)
f = JTF
(2.2)
The inverse of the Jacobian matrix can be found through the observation of the relationship between the force vector at the mobile
plate and the forces applied by six actuators, and expressed as the
following:
10 1 0
1 0 0
0
2L/3
0 1 0
-L/3
-2L/3
0
?;? 1
c;1
49
for high force generation and the high energy density, which makes
Terfenol-D devices quite compact [4].
Several version of Terfenol-D actuators have been built by a few
companies, but the maximum linear displacement has been limited to
f 6 3 . 6 microns (f2.5 thousands of an inch). We decided, at the outset
of this effort, to attempt to build a Terfenol-D actuator which had
greater displacement. Our goal was f 1 2 7 micron ( f 5 thousandths
of an inch). Another goal was to implement an actuator which had
both a flux monitoring capability and and a direct strain measuring
capability. This was not available from any known commercial source
of Terfenol actuators. Our Terfenol-D actuator design, as shown in
Figure 4, incorporates an instrumented Terfenol-D rod, an excitation
coil to provide the magnetic driving field, a permanent magnet
assembly to provide a magnetic bias field, and a mechanical pre-load
mechanism.
The instrumented Terfenol-D element we used has a diameter
of 0.00635 m. (1/4 inch), and is approximately 0.1524 m. (6 in.)
long. The instrumentation we developed consists of a number of flux
pickup coils and strain gauges. The flux pickup coil was used to
directly measure the flux change in the Terfenol, and the strain gauge
bridge directly measures the elongation of the rod. The Terfenol rod
elongates in response to a magnetic field generated by the signal
coil. The wire size is selected to match the actuator impedance to the
drive electronics. It can generate 500 Oersteds per ampere. Magnetic
bias is required to set a magnetostriction inside the Terfenol-D rod
caused by coils in the center of its linear operating range, so that an
alternating current applied to the signal coil will cause the actuator
to alternately extend and contract depending on the direction of the
generated by the signal 'Oil* Permanent magnets are
used in our actuators for the magnetic bias. Extensive design and
manufacturing modifications results in actuators which works well.
We achieved a travel o f f 127 microns ( f 5 thousandths of an inch).
The output stroke V.S. input current characteristics of Terfenol-D
actuators are non-linear and exhibit hysteresis, due to the material
property of Terfenol-D. However, in a small range of motion, the
characteristics can be approximated by:
6=Ku
(3.1)
FILTERALGORITHM
IV. A ROBUSTADAPTIVE
FOR ACTIVE VIBRATION CONTROL
Design of an active vibration control system is a challenging
task because of the large number of structural modes involved
in the bandwidth of interest and the large amount of uncertainty
involved. In our application, the problem is exacerbated by the fact
that the dynamic behavior of these systems will hardly be known
precisely during design, and identification of system model will
provide only limited information because of uncertainties such as
dynamic perturbations, transient thermal states, parameters drift, etc.
Furthermore, the actuators have nonlinear and unknown parameters.
All these facts raises concem as to robustness and stability of active
control systems.
There have been a number of active vibration control design
approaches proposed in recent years. The control design of most
of these approaches relies on the model information to predict the
structure's dynamic response. In reality, a perfect model of a lightly
damped space structure has an infinite number of lightly damped
modes. Consequently, higher order modes are truncated from the
model in order to perform effective control design. The model-based
50
e ( k ) = Y I ( ~ ) yz(k) = &(z)u(k)
+ Pz(z)f(k)
(4.1)
f(k) = H ( z ) u ( k ) .
(4.2)
f(k) = -P;(z)P,(z)u(k).
(4.3)
channel with FIR filter. However the same idea applies to multiple
channel systems or the system which adopts IIR filter.
To construct an FIR filter which approximates the behavior of
Pz(z),we define:
m
Pz(t)~(k)
= V T ( k ) X ( k )= C ~ i ( k ) z (-kj)
(4.6)
r=O
Pz(z)f(k)= C W i ( k ) W T ( k- i ) U ( k - i )
i=O
m
= C C w i ( k ) w j ( k - i ) ~ ( l c - i - j ) (4.7)
i=O j=O
- m)lT.Thus
e ( & )=Pl(%)u(k)+C~vi(k)wj(k-~)u(k-z- j ) .
(4.8)
i=O j = O
(4.1 1)
YZ(k) = H ( 4 V T ( k ) U ( k ) .
(4.12)
s(k)=V T ( W ( k )
(4.13)
Define
+H(z)g(k)
= P(z)u(k) + C w j ( k ) g ( k -j>.
= Pl(z)u(k)
(4.14)
j=O
f ( k ) = C W j ( k ) U ( k -j) = W T ( k ) U ( k )
(4.4)
j=O
Therefore
(4.15)
e ( k ) = A ( z ) ~ ( k ) Pz(z)WT(k)U(k)
(4.5)
(4.16)
V ( k + 1) = V ( k )- P p e ( k ) F ( k )
W ( k + 1) = W ( k )- 2 p e ( k ) G ( k )
where G ( k ) = [ g ( k ) [ g ( k- 1 ) . . . g ( k - .)IT.
(4.17)
(4.18)
Equations (4.17) and (4.18) provide explicit adaptation laws for the
filters weight vectors W and V. The computation flow is shown in
the gray block of Figure 5, where filter 2 adapts its weights V based
on error e and secondry control command f, while the function of
filter 2 is to provide a filtered U signal g for the adaptation of
filter 1 based on input history. Filter 1 acts as feedforward controller
which generates an active control command f from input disturbance
U, and its weights W are adjusted according to the output error e and
filtered-u signal g.
The proposed algorithm estimates and tracks both the feedforward
controller transfer function and the secondaq channel transfer function in an on-line fashion. Therefore it can deal with severe system
and environmental changes and maintain acceptable performance. In
our real-time experiments, we deliberately changed the system parameters such as mass, and the proposed algorithm recovered quickly
from the unexpected change and readjusted to the performance before
the change.
51
52
Fig. A l .
fz
= f 2 sin B
fv=f1+f2cos~
f = J Z Z j == Jf; + f;
+ 2 f i f i cos B.
The value of p where f reach its minimal value can be found through
function g;
g = p*
+ 2p%osB + 1
= 2p
+ 2 c ~ = ~o e
t
resulting:
p = -cos@
therefore,
min
fp=-
coee = f i sin 8.
On the other hand, f may reach its maximum value at the boundary
of p , which is either 1 or -1. We have:
f;:
= fld-.
APPENDIX 1 DISCUSSION
OF UNIFORMITY OF CONTROL AUTHORITY
Lets consider first a two dimension case with two actuators.
Without lossing the generality, assume actuator 1 aligns with X axis,
and the angle between actuator 1 and actuator 2 is B , which ranges
from - x / 2 to ~ / 2The
. f 1 and j 2 are the force generated by actuator
1 and actuator 2, respectively. We will first find the maximum and
minimum control authorities of this configuration at a given B angle.
Then we will show that the ratio of the minimum and maximum (i.e.,
uniformity) reaches maximum when 0 equals to 7r/2.
ACKNOWLEDGMENT
The authors would like to extend their thanks to the sponsors
of this work from National Aeronautics and Space Administration
(NASA) under the grant NAS7-1134, and NAS7-1198. Special thanks
go to Mr. John Garba, Mr. Ben Wada, and their colleagues at
the Jet Propulsion Laboratory for their cooporation and numerous
suggestions.
REFERENCES
53