, DRAFT

_ .. 12/19/90

c o,,v
F-¢,,o -/
q ATTITUDE CONTROL OF A SPINNING ROCKET SAND--90-3230C
VIA THRUST VECTORING
DE91 006267

John E. White

Sandia Natlonal Laboratories

Albuquerque, NM 87185

Two controllers are developed to provide attitude

control of _ spinning rocket that has a thrust vectoring
capabillty. The first controller has a slngle-input/single-
outputdesign that ignores the gyroscopic coupling between
the control channels. The second controller has a multi-
input/multi-output structur_that is specifically intended
to account for the gyroscopic coupling effects. A perfor-
mance comparison between the two approaches is conduct_d for
a range of roll rates. Each controller is tested for the
abillty to track step commands, and for the amount of
_oupling impurity. Both controllers are developed via a
llnear-quadratic-regulator synthesis procedure, which is
motivated by the multi-lnput/multl-output nature of second
controller. Time responses and a singular value analysis
are used to evaluate controller performance
i

I. Introduction .

This paper describes thl development and comparison

of two controllers that are designed to provide attitude

=
control of a spinning rocket that i_ equipped with thrust

vector control. Both controllers are evaluated as to their

__ ability _o track step commands, and for the ability to

provide adequate decouplingbetween the pitch and yaw

control channels. The dynamic coupling between the two

channels is due to gyroscoplc effects when the rocket has a

non-zero roll rate. This problem could be of interest for

several reasons. For a single nozzle rocket design there

may be no direct control of roll _ttltude and some roll rate

- may be introduced by disturbances. In the case wher_ roll

DISTRIBUTION OF THIS DOCUMENT IS UNLIMITEEU

MASIER
 2

control is available, a failure in the roll control channel

might introduce a significant roll rate. Finally, it might

be possible to lower the cost and weight of delivering a

spln-stabilized payload by spinnlng-up the booster prior to

motor burn-out and payload release.

The first controller is basically a single-input/single-

output (SISO) design that ignores the gyroscopic coupling

effects and attempts to provide adequate control as if the

channels are dynamlcally uncoupled. The second controller

attempts to account for the gyroscopic coupllng effects via

a multl-input/multi-output (NIMO) design approach. A llnear-

quadratic-regulator (LQR) control system synthesis procedure

is used to design both controllers. This design technique

is motivated by the MINO nature of the second controller,

since LQR is a matrix procedure that inherently produces the

proper loop inter-connections. Both controllers are shown

to be capable of providing steady state tracking (for no CG

offsets), although the decoupllng characteristics of the

MIMO controller are slgnificantly better than those of the

SISO design even for relatively modest roll rates. A gain

adjustment technique for the MIMD controller is shown to

&pproximately maintain the closed-loop transfer function of

a baseline design as a function of actual roll rate, and

reduces the MIMO controller to that of the SISO controller

as roll rate approaches zero. The stabillty margins asso-

ciated with the SISO controller are shown to be decreased

relative to the MINO controller for non-zero roll rates.

=

i
 3

The organization of this paper is as follows. Section

II derives the nonlinear dynamical equations of motion,

which are then simplified to produce the linear dynamics

that are the basis for the control system design. Section

III develops the state spaces that determine the controller

structures, and discusses the LQR synthesis procedure.

Section IV contains a brief description of a singular value

analysis that is used to evaluate controller stabillty

margins and robustness to parameter uncertainty. Section V

discusses the performance of the controllers as determined

from a simulation study. Conclusions are presented in

Section VI.

II. System Dynamics

A. Nonlinear Dynamics

This section derives the nonlinear equations of motion

for a spinning spacecraft system. The vehicle and its body

fixed reference frame are shown in Fig. 1, and assumed

physical data are given in Table 1. An inertial reference

frame is defined to be coincident with the body frame for a

zero roll angle. The inertial frame is preferred from a

control design perspective [1], since controller commands

are most naturally expressed in this frame. The equations

of motion used by the simulation are written with respect to

the body frame [2], however, since the control torques are
 applied in this frame. The vehicle is assumed to be

spinning about the body X-axis. Two perturbation masses

with radial displacements y and z along the body axes Y

and Z will be assumed in this analysis as a means for

introducing radial center of mass offsets into the

simulation. The center of mass of the system, Including

control masses, is located at point A for y-z-0. Point O is

the new location of the center of mass when the control

masses are located to a position off of the vehicle

centerllne. The body axes are located with respect to point

A, and the vehicle is assumed to be nominally symmetric with

mass moment of inertia of Ix about the X-axls and equal

inertias of I about the Y and Z axes. The perturbation

masses are assumed to be of equal mass, m.

The nonlinear equations of motion can be derived from

- where the left side of the equation represents the sum of

_= the moments about point A, and the right side is the time

m rate of change of angular momentum. The only external

-

-
torques assumed to be acting on the system are due to the

5°
rocket thrust vector control moments, which are expressed as

- MA - ( T i sln(6y)) _ + ( T i sin( 6 z 11 _ where T is the

!
thrust of the motor, I is the moment arm between the nozzle

and the center of mass, and 6 is the deflection of the

nozzle. The location of the vehicle center of mass with

respect to point A is given by

1
 i
' 5
l
|

i
! " ( _y ) 3 + ( _z ) _ (2)
i

i where I, _, _ are unit vectors along the X, ¥, Z axes, and p

!

0 iS the mass fraction m/m T . The acceleration of point A is

-"-# [ ( Z_B-y_B + ypBqH + zPHrB) 1

+ ( _ Z_SB- yp2 + zqBrB_ Yr2B)

+,c YqB- B - ) c3)

where the angular velocity with respect to the rotating body

i frame is _ - pB _ + qB_ + rB_ , and _ = _ - 0o Flnally

]_A X mta A - -pm [ ( y21bB+ y2qBr B- yz(_B+_S)

+ Z2_B+ yzp 2 - z2qBrB + yzr 2) 1

- + ( Z2_B - yZ_B+ YzPBqB + z2pBrB )

'i - ( ylq B- y2_B+ y2pBqB+ yzPBr B)

The angular momentum of the vehicle with respect to A is

-i _A " [ _ _ _ ] IA_ - m ( _m X _m ) (4)

(5) .

|. where rm= x_ + y_ + z_ and _m = _ + _ = 0, and

J
Ixx -mxy -mxz
_- 1A - -mxy I + mz 2 0 (6)
i - -mxz 0 I + my 2

i Hence _A" [(Ixx+my2+mz2)pB- (mxYlqB- (nucz)rB] 1

+ [(-mxy)PB+ (I+mz2)qB ]

+ [ (-_xz) pB+ ( I+my2 ) rBl _ (7 )

i
,lm

I
 The derivative of _A is written as

d hA A+ _ X hA ) (8)

where the two teI-me o'l the right hand side are given by

_A" [ Ix_B + my2_B + mZ2_B" mXY_B-mXZ_s] _

+ [ I_ B- mxYlbB+ 2mX2_B+ mz2_ Bl

+ ( If B- mXZ_s+ 2mX2fs+ my2rBl

(9)

(_ X _A ) -m [ (y2_ z2)qBrB+ XypBrB - xzpBqB ] i

'2 2
+ [ (Ix-I)PBrB + m ((z2-2x2)PBrB - xyqBr B- xz(rB+P B) ) ]
2 2
+ [(I-Ix)pBqB + m ((2x2-y 2)pBqB - xy(pB-qB) - xzqBr B)]
(i0)

Substitution of (9) and (10) into (4), and then (4) and (8)

into (i) produces

Cp -mxy -mxz _B 0 -myz myz _B

-mxy Cq #myz _B " mxz 0 mxz qB

-mxz #myz Cr fB mxy -mxy 0 -_

i (l
+ #myz mxy (I-I x} qBrB
-ro(l-#) z
(Ix-I) -mxz -#myz rBP B
+su(1-_ )yi

I
where Cp - Ixx + m(1-#)(y2+z2)
+

I0011 T1 s
0

, Cq -
0
T1 s

I
sin(Sy)
sin(6z)

+ m(1-#)z 2 ,
(11)

2 12 2 2
Cr - I . m(1-#ly 2, and Is + (#y) + (#z) . Nonlinear

differentlal equations in PB, qB, and r B are obtained by

inverting the matrix on the left hand side of (11) and

premultlplylng the right hand side. This inverse is

CqCr-# 2m2y 2 z2 CrmXy-#m2 xyz 2 -#m2xy 2Z+CqmXZ

1 Crmxy_#m2xyz 2 CpCr_m2x2z2 _ Cp#myz + m ._.

x 2yz (12)

-#m2xy 2Z+CqmXZ -Cp#myz+m2x2yz CpCq-m2x2y 2

where D = CpCqC r + m3#x2y2z 2 - Cp#2m2y2z2 - Cqm2X2Z 2.

The equations of motion with respect to the inertlal

frame are needed for the purposes of control system design.

A reasonable assumption is that y-z-0. The angular velocity

with respect to the inertial frame is _i = (pB+p)_ + q_ + r_.

Equation (1) can expanded with respect to the inertial frame

to produce
m.

MA - [_I_Iki
-_ ] IA_I_ + ( _I X [_I _ IkI
]* IA_I ) (13)
 B. Linear Dynamics

The linear dynamical equations of motion for the spinning

rocket with respect to a non-rotating (inertial) command

frame (X I,YI,ZI) can be deduced from the nonlinear equations

(13) from Section A. The simplifying assumptions are:

(I) rocket body spin axis, X, coincides with XI (i.e., p=0),

(2) the body x-axis is an axis of symmetry,

(3) the rocket has a spin rate of PB relative to XI,

(4) the Inertlal frame is not rotating, and

(5) all external torques are applied by thrust vectoring.

The nonlinear eqations of (13) can now be ceduced to

- 0 (14)

.- ( ____X) PBr + ( iT!l ) sin(&y) (15)

= + ( ___x) pB q + ( TII ) sin(6z ) (16)

The right hand sides of (15) and (16) represent the gyro-

scopic coupling terms and the control moments applied by the

thrust vector control system. Since the" deflection angles

_ are limited to a maximum of 3 degrees, (15) and (16) can

further be reduced with the as_umptlon that sin(6) - 6.

A linear, second-order actuator model can be defined as

i " ; 6"Idt (17)

-- i - H16i + H1ui (18)
where the i subscript is replaced with y or z depending

on which actuator is under consideration. The control

variables become the actuator position inputs of u i.

"l

%
 III. State Space Formulation And LQR Synthesis

A. State Spaces For LQR Controller synthesis

A general characteristicof control systems is that of

driving some s_teady-state tracking error to zero. It is

also generally considered desirable that the zero steady

state tracking error be insensitive to unmodelled system

dynamics. The system tracking errors are, therefore, natural

state variables for the LQR synthesis procedure [3]. The

two tracking errors in the spinning rocket problem are '_

Es - 0C - e (19)

E_ = ¢c - _ (20)
where the variables with C subscripts are assumed here to be

constant step input commands, and the state varlables (0,_)

are measured pitch and yaw angles. The differential

equations for the tracking error states can be written as

E0 " - q (21)

_ = - r (22)
The state space used with the LQR synthesis procedure is

now determined separately for the SISO and MINO contr_T]ers.

| First consider the SISO pitch angle control problem. The

-

SISO yaw angle controller will have an identical compensator

i structure, and
be symmetrical. theThis
samecontroller
gains if the vehicle
isdesigned is assumed
to track a to
step
I

input pitch angle command, but must be able to accomodate

I
the unmodeled effects neglected in the simplified linear
i
model if this objective is to be successful. The vehicle is
|
|

i
 10

assumed for design purposes to have no radial center-of-

gravity (CG) offsets• The gyroscopic coupling terms are

also neglected, so that (15) reduces to

- Ny6y (23)

where My can be defined from (15). A consistent state space

for the SISO pitch angle controller can be obtained by

combining (21) and (23) to produce

Eo " - q (24)
= My&y (25)

In general the nozzle deflections will not be zero in steady

state except for vehicles that are perfectly symmetric about

the body X-axls. The above state space will produce a

proportional-derivative (PD) compensator structure that will

provide steady state tracking only for symmetrical vehicles.

This formulation can be interpreted as having an unbounded

LQR performance index, since the nozzle deflection state is

| not zero in steady state. Integral compensation must be

included, if steady-state tracking is desired for radial C_

| offsets. This is generally accomplished via a steering loop

that is in _urn commanded by the guidance law. Simce no
|
guidance/steering is included in this study, integral

compensation compensators. accom-

i is added to the This is

|| plished by differentlatlngthe state space of (24-25) and

•. adding a state variable so that the error variable is

retained. This step is equivalent to adding an integrator

I
|

_
 11

to the compensator, since the new control is the derivative

of the desired control and an integrator must be inserted

between the compensator and the actuator for implementation

purposes to provide the proper actuator input slgnal. This

integrator can then be pulled through the summing junctions

of the compensator to produce Integral compensation. The

LQR compatible state space is

%e o z o o Ee 0
oo J

EO_ o o -z o _e o
= o o o My ,_ + o _y
_y 0 0 0 -H 6y H (26)

where a first order actuator approximation is used, since

the higher order actuator state feedbacks are not measured.

Alternatively, the second order model of (17'18) could be _

used, but the resulting gains on the actuator rate state are

set to zero. This state space is controllable, does not

include any input command dependent terms, and all of the

states an_ controls are zero in steady s%ate for step input

' commands. This state space is, therefore, suitable for LQR

synthesis. The resulting proportlonal-integral-derlvative

_PID) compensator structure is shown in Figure 2, where the

derivative compensation is Implemented as rate feedback.

The MINe controller LQR compatlble state space can be

determined in a manner similar to that used for the SISO

, case. The gyroscopic terms are, however, retained in the

linear dynamics. (21) and (22) and their derivatives are

 12

combined with the derivatives of (15) and (16) to produce a

consistent state space of the form _ - A x + By, where

xT" [ E0' E0' _' E_, E%, f, 6y, 6z] T and vT - [ _y , _z ] '
where the A and B matrices are defined, respectively, as

0 1 0 0 0 0 0 0

0 0 -i 0 0 0 0 0

0 0 0 0 0 Mr 0 0
A m (27a)
0 0 0 0 1 0 0 0
0 0 0 0 0 -i 0 0

0 _ 0 Nq 0 0 0 0 0
0 0 0 0 0 0 H 0
0 0 0 0 0 0 0 H

m 0 0 0 0 00 0 H0 H0 (27b)
BT I
L
0 0 0 0 0 1 T
...

This state space is controllable, has no input command

dependent terms, and has zero steady state values for all

states and controls for step commands. The resulting MIMO

compensator has a proportional-integral-derivative (PI)

structure. The controllability of the N_MO state space must

be evaluated after the addition of any integrator states to

avoid the introduction of uncontrollable modes [3]. The

resulting compensator structure is shown in Figure 3.

Since the control p_oblem has been formulated with

respect to the inertial reference frame, all feedback to the

controllers must be expressed in this frame. The control

inputs to the actuators must then be transformed back into

the body frame.

,
Both steps require knowledge of roll angle.
 13
J+

B. Linear Quadratio _e_'...a_or Procedure

The L_R synthesis problem in that of determining the

oontrol that _Inlesizes the performance index

J _ I ( xTQ x + rTR v) dt (28)

where R > 0 a_d Q > 0, subject to the dynamic constraints

_, " A X + B V (29)

q where the A and R matrices f/_r the spinning rocket thrust

vector control loops are defined symbollically in part A oZ

- this section. The solution to this optimization problem is

| a linear controller of the form

J

_m

|- v - - ( R-IBTp )x - -G x (30)
0|
where P is the algebraic matrix Riccati equation solution
i

•| [5,6]. _.'_he constant gain matrix, G, can be determined from

any one of several computer-aided control system design

_

software packaqes which are currently available to solve the

_. LO_ optimiration problem. Although the above optimization

problem ks wtated in continuous time, the simulation test

qains for the spinninq rocke_ thrust vector controllers are

_ based upon deter_inlng the discrete form of (29) and the

_

_ J
assoclated discrete gains for a particular sample rate.

- The LQR state weighting matrix for both the SZSO and

- MIMO compensators is chosen so as to satisfy several design

!

_ criteria. Xn each case Q is chosen to position the closed-

loop poles such that they provide acceptable transient

: syste_ parameter variation insensitivity. The R matrix is

sm

qm

i
 14

chosen as an identity matrix for the MIMO case since there

_e no physical reason to weight one control overt he other.

Adequate decoupling between the pitchand yaw channel_

is also a primary design requirement. This is accomplished

for the MIMO case primarily with the choice of the weighting

on the rate states, but is also a lesser _unction of the

error and integral of the error s_e weights since the

llnear dynamics are a very simplified form of the actual

system dynamics. The weights on the rate states ml l_ add

adequate damping to the open loop poles along the J_-axis,

while maintaining a system bandwidthsufficlently slower

than the actuator natural frequency. Sensitivity to

parameter variations will occur if the actuator bandwidth

is not at least a factor of 4 greater than the system

bandwidth. The selected system weighting matrice_ are

shown in Table 2.

The degree of coupling between t"' control channels is

dlrectly related to the vehlcle spin rate, as indicated by

(15) and (16). If vehicle inertia prope_ :ies are

i

assumed to be constant, the cross-channel gains in the MIMO

compensator can be shown to be directly related to the spin

rate of the rocket. These gains can be adjusted with a

scale factor that is a ratio of the actual to design spln

rate to approximately maintain the closed-loop transfer

" function of the baseline design. This procedure produces a

the roll rate approaches zero.

mj ' ii, , ' ' r'

 15

IV. Singular Value Analysis

The robustness of the spinning rocket thrust vector

controlleEs have been inv astigated with the use of singular

value anal_,3is. Singular value analysis is analogous to the

SISO gain and phase margin measures of the closer._ss of the

Nyquis_ plot to the (-1,0) point, in that the measure of

this distance for MIMO systems is in terms of the

singularity of a matrix. The LQR has been shown in [7-9] to

have useful robustness properties. For a diagonal error

model, the LQR has simultaneously in each feedback loop a

guaranteed minimum gain margin of ( -6, ® ) dB and a

guaranteed minimum phase margin of _+60 degrees. Since the

higher-order actuator states are not available as feedbacks,

a measure of the gain and pha_e margins of the proposed

des_.gns in required. The procedures of [10] are applicable,

given the singular values of the return difference matrix

D(s) - [ I + X(s) G(s) ] (31)

where K(s) is the compensator transfer matrix and G(s) is

the plant transfer matrix given by Gp(s) = C(sI-A)-IB such

that K(S)Gp(S) - G(sI-A)-IB. These gain and phase margin

results are associated with the nominal system. Singular

value analysis can also be useful in assessing the robust-

ness of the system with respect to parameter variations and

neglected dynamics. The matrix of most interest here is the

inverse return difference matrix

, -i
; T(S) = Z + (X(S) G(S) ) (32)
M

I
t
 lC

The central robustness relationship [11] is that

| a
m (T(s) ) > _ (L(s) :) for all s - J_ (33)

where _ is the smallest singular value, and _ is the largest

singular value. L(s) is _a multipllcative error model, since

the perturbed plant, G* is expressed as

p'

Gp(S) -Gp(S) [ I + L(s) ] (34)

If condition (33) is satisfied, then stability is

guaranteed. This condition is conservative, and failure to

meet this condit_.on does not necessarily imply system

ins tabillty. The recent literature [12] has described

efforts at developing less conservative stability tests.

The construction of _he error models is important in order

to evaluate control system robustness. The error models of

most concern here are those associated with the important

system parameters in the A and B matrices. The error model

associated with parameter variations can be derived with

the use of a matrix identity. Define t(s) - (si-A) -1. A

parameter variation can alter both the A and B matrices.

These perturbations are denoted here by AA and AB,

respectively. An error model can be derived [3] as

L(s) - GplCt [ AB + AA(I+tAA)-ltB - AA(I+tAA)-ItAB ] (35)

This error model has been used to evaluate the robustness of

-
both controllers, as is discussed in the next section.
--.

,in
1

• ,
 17

V. System Performance

The performance of the SISO and MIMO controllers have

been evaluated from a slngular value analysls, and from time

response behavior against adynamic simulation based on

(11)-(12). Controller gains and the associated weighting

matrices and system eigenvalues are given in Table 2.

Simulated transient responses are given in Figs. 4-13. The

despun system response for a 10 degree yaw command is shown

in Figs 4-5. The time response of the two controllers is

nearly identical for this situation. Figs. 6-7, 8-9, 10-11,

and 12-13 show the behavior for roll rates of 0.25, 0.5,

1.0, and 2.0 Hz, respectively. These plots are polar plots

of pitch angle versus yaw angle to emphasize the coupllng

between the control channels. In each case the decoupllng

performance of the MIMO controller is better than that of

the SISO controller. This difference in performance is

particularly obvious if the roll rate is increased to a

level that stresses the frequency _esporJe capability of the

• actuator. For the actuator parameters given in Table I,

the SISO controller will go unstable for a roll rate between

2 and 3 Hz, while the MIMO controller Is stable out to 5 Hz.

L

If the actuator is alower than that used for these results,

then the difference in decoupling performance is exaggerated

J
at lower frequencies. Off-nominal responses due to para-

meter variation in the nozzle deflectlon effectiveness are

also shown in Figs. 4-13. The effectiveness parameters are

._ -- _h..v4_I r_ In_A_4nn. which can vary

i
 G

18

i
'
i signlficantlyduring the burn of a large rocket. The actual
|

i vehicle roll rate for scaling the baselin_ NINO gains is

i!
i another parameter of interest, but reasonable variations in

the measured roll rate does not appreciably degrade perfor-

mance. The off-nominal transients include variations of

-TBD% in the effectiveness parameters for the NINO con-

troller, and -TBD% for the SISO controller.

A singular value plot of the return difference matrix

(31) for the MIMO controller is given in Fig. 14. This plot

shows a minimum of -TBD dB. since an LQR design with full

state feedback has a minimum of 0.0 dB, there is a penalty

i in stability margin for the lack of full state feedback (no

actuator rate feedback). A similar plot is shown in Fig. 15

! for the SISO controller. This plot shows a minimum of -TBD

_ ..

dB. The gain and phase margins can be determined from [I01
,|
q as a gain margin of ( -TBD, +TBD ) dB and a phase margin of

±TBD degrees for the MIMO controller, and ( -TBD, +TBD) and

±TBD degrees for the SISO controller. Fig. 16 shows the

singular values of the inverse return difference matrix (32)

!_ for the NINO controller, plotted against the error model

1 (351 for the parameter variations of -TBD% in the

effectiveness parameters. Fig. 17 shows a similar plot f_r

R

i the SISO controller for a variation of -TBD%. Since

:I_ _(L} < _(T} , _tabilit_ is guaranteed for these variations.

ii Simulations and stability analysis bo_hshow that the MIMO

controller is more robust to variations in this parameter.

l
 19

VI. Conclusions

Two controllers for the thrust vector control of a

spinning rocket are described and compared. Oneof the

controllers is a single-input/single-output (SISO) design

that assumes that the vehicl_ is not spinning, so that there

is no gyroscopic coupling between the pitch and yaw control

channels. The second controller formulationincludes terms

that account for the gyroscopic coupling when the rocket

motor is spinning. The multiple-input/multiple-output

(MIMO) nature of the latter case has motivated the use of a

linear-quadratic-regulator (LQR) synthesis procedure. The

choice of the state space upon which the LQR synthesis is to

be applied is analogous to the selection of thecompensator

structure in classical control theory. The state spaces are

selected to ensure zero steady-state tracking for constant

angle step commands, by the i addition of states which are

equivalent to integrators in the compensator. Both

controllers have a proportional-integralyderivative

structure. Integral compensation was included for the

convenience of this study. A guidance/steering outer loop

would normally provide the same effect, such that integral

oompensationwould not be required in the control loops.

The weighting matrices selection for the LQ_ performance

|
index is analogous to the choice of gains in a classic ll
I
approach. The performance index weights are chosen here by
m

I
a trial and error approach to achieve the primary goal of
q
the compensator, which is to provide adequate tracking of
W
m
J
 L

20

pitch and yaw attitude commands. For aspinning vehicle,

i

adequate decoupling between the two channels is also

desirable. The quality of the decoupling is a function of

the vehicle inertias, spin rate, and actuato_ capabilities,

as well as the structure and gains of the compensator. A

simple technique has been proposed and tested for adjustment

of the MIMO controller gains as a function of the actual

spin rate ofthe vehicle to properly account for the

gyroscoplc coupling between the control channels. As the

roll rate decreases to zero, the MIMO compensator structure

reduces to the uncoupled SISO structure.

The time response behavior of the two control systems

has been tested in a three-degree-of-freedom, nonlinear

simulation. The performance of the SISO controller,

although generally adequate, is definltely degraded even for

- moderate roll rates when compared with the MIMO compensator.

The differences between the two designs are particularly

apparent as the spin rate is increased. The most serious

drawback occurs as the spln rate is increased to the limit

of the ability of the actuators to respond, in that the SISO

t

controller will go unstable well before the MIMO controller.

This is an indlcatlonof reduced stabilty margins for the

i been
SISO confirmed
design via a singular
relative value
to the MIMO stability
design, a factanalysls.
which has
ii
ii
.I
I
m
m

i
!
I
J
i
!
i
I
 r, + I

r 21 +

REFERENCES

i_i] Fortescue, P.W., and Belo, E.M., "Control Decoupling

Analysis for Gyroscopic Effects in Rolling Missiles,"
Journal of Guidance, Control, and Dynamics, Vol. 12, No+ 6,
Nov.-Dec., 1989, pp. 798-805.

[2] Kaplan, Marshall +H., Modern Spacecraft Dynamics and

+Control++ John Wiley and Sons, New York, 1976, pp. 173-180.

[3] Athans, M., "On the Design of P-Z-D Controllers Using+

Optimal Linear Regulator Theory," Automatica, Vol. 7, Sept.
1971, pp. 643-647.

[4] Speyer, J.L., White, J.E., Douglas, R., and Hull, D.G.,
"Multi-Input/Multi-Output Controller Design for Longitudinal
Decoupled Aircraft Motion," Journal of Guidance, Control,
and Dynamics, Vol. 7, No. 6, Nov.-Dec. 1984, pp. 695-702.

[5] Brockett, R., Finite Dimensional Linear Systems, John

Wiley and Sons, New York, 1970.

[6] Bryson, A. and Ho, Y.-C., Applied Optimal Control,

Blaisdell, Waltham, Mass., 1969.

[7] Sain, M., ed., "Special Issue on Multivariable

Control," IEEE Transactions on Automatic Control, Vol.
AC-26, Feb. 1981.

[8] Safonov, M.G., and Athans, M., "Gain and Phase Margin
for Multiloop LQG Regulators, " IEEE Transactions on
Automatic Control, Vol. AC-22, April 1977, pp. 173-179.

[9] Lehtomaki, N., Sande11, N., and Athans, M., "Robustness

Results in Linear-Quadratlc Gaussian Based Multivariable
ControlDesign," IEEE Transactions on Automatic Control,
Vol. AC-26, Feb. 1981, pp. 75-92.

[10] Mukhopadhyay, V. and Newsom, J., "A Multiloop

Stability Margin Study Using Matrix Singular Values",
Journal of Guidance, Control, and Dynamics, Vol. 7, No. 5,
Sept.-Oct., 1984, pp_ 582-587.

[11] Doyle, J. and Stein, G., "Multivariable Feedback

Design: Concepts for a Classical/Modern Synthesis," IEEE
Transactions on Automatic Control, Vol. AC-26, Feb. 1981,
pp. 4-16.

[12] de Gaston, R.R.E., and Safonov, M.G., "Exact

Calculation Of The Multiloop StabilityMargin," IEEE
Transactions On Automatic Control, Vol. AC-33, Feb. 1988,

i
m pp. 156-171.
m
 mt- zooo Ibm, m- 10 ibm
Ix" _400 ib m ft2, I - 5500 Ib m ft2

R - 2.25 ft, 1 - -3.2 ft, L- 19.0 ft

PB " 6.2832 rad/sec (Baseline roll rate)

Actuator Model: fn - i0 Hz, _ - 0.6

TABLE 2: CONTROLLER PARAMETERS

SISO Controller

Gains: 0.0316 0.0586 -0.0385 0.0752

Diagonal Weights: Q(I, I)=Q(2,2)=0. 001,

Q(3,3)=0.0005, Q(4,4)=0.0

Open-Loop Eigenvalues: 0.0

0.0 + jl.888
| "62.8

-z.zs99_+jo.s0 5
-62.7354
Closed-Loop Eigenvalues: -2.4079
a MIMO Controller

i Gains:
• 0.0295 0.0554 -0.0378 0.0113 0.0212 0.0006 0.0739 0.0000

Diagonal Weights: Q(I, I)-Q(2,2)-Q(4,4)-Q(5,5)-0.001,

i -0.0113 -0.0212 -0.0006 0.0295
Q(3,3),-Q 0.0554
(6,6)-0. 0005,-0.0378 0.0000
Q(7,7)-Q(8,8)-O. 0.0739
0

Open-Loop Eigenvalues: 0.0

0.0
0.o+_jz.sss5
-62.8
-62.8

." Closed-Loop Eigenvalues: -0.9419 +_ J0. 6109

| i -1.0239 :i: j0.3760
-_ -2 7371 + Jl 8267
W-- -62.7355 + J0.0033
l

|
l

J
lr, •

• i /
,
w , u

_,me'

..
_.J
Zrr o
_-_W 0

j ....
p-Q
mn,,
_--
UZ
OG
0_f(_.) o

Z>

ZO
Z ,,
( _
°
o
,

0"OZ 0"91 O'Zl 00'9 O0'IT 00"0

9]OlSd
 O
O
t i ii , '

o O
Z o
o .......... ,.5 _.
I--
_E ''
....J tn

Zn/ o
_..4LI..I O I_
u")d j i ,,,
' ---_ up
t-E)
I..LIrv'
_I---
UZ
(DO
o
OU .... '........ --ai
z> \
ZO
0.. ,-,

tB i i., , , .li,. t i i , i , i i _._ 0

0"0Z 0'9t 0'ZI .,P'S""_J"t _ _

00'8
! ,
00 _, 00"0

_I.'IU
Ibcl I

ii ii li ul
I
i.
o
0

0
 i *o _'

_
- !
I
i
 0

........ _
c_
N

z __. °
_ o

_..I ff.I _

_n,, b '
_--4I.LI oLD ..
O_._l OW 0
.s _...... n -
WC_ CO UJ
__ . n
LJZ d_

Z
ZO
=i O_",
,, .

:I O0 Z OZ I O_ 0 OI,'O- OZ'I- O0'Z-

(030) 03011-II

i, i ,, , ,, _

q
n , ,.... ,,
CD
6
 0

-(',d
c_

_.(.D

0
z o ,4
o ,., _ .d _
I---I _f uJ
_.-,I
A _:

i _ _ °°-.
oo

°
. uZ _ / ' ' -o60
a oo _-
i " 0
EEL) _ LI.

l_ ,--,I--
Z
tl zo
i O_ _-,4
-.--Na 0

_ " "0 ,

-. 0_"0 OZ'O 00'0 OZ'O- 0_'0- 09"0-

ii
il (O3O)0301Hl
!_

q
ii

alli

i
" I,,
 DISCLAIMER

This report was prepared as an account of work sponsoredby an agency of the United States

i employees,
Government. makes any warranty,
Neither the United express
States or implied, ornor
Government assumes any legal
any agency liability
thereof, or responsi-
nor any of their
• bility for the accuracy, completeness, or usefulness of any information, apparatus, product, or
II process disclos=_,d,or represents that its use would not infringe privately owned rights. Refer-
j ence herein to any specific commercial product, process,or service by trade name, trademark,
=
_!1 manufacturer, or otherwise does not necessarily constitute or imply its endorsement, recom-
i mendation, or favoring by the United States Government or any agency thereof. The views
i 1 and opinions of authors expressed herein do not necessarily state or reflect those of the
I United States Government or any agency thereof.

i
l

=
,m