Sie sind auf Seite 1von 33
. NASA TP . 1118 c.1 3 NASA Technical Paper 1118 - . Eigenvalue/Eigenvector Assignment
.
NASA
TP
.
1118
c.1
3
NASA Technical Paper 1118
-
.
Eigenvalue/Eigenvector
Assignment
S. Srinathkumar
PEB~UARY1978
NASA

NASA Technical Paper 1118

TECH LIBRARY KAFB, NW

IllllllH1Iullllllll11111illIlllIIIIll

0134489

Eigenvalue/Eigenvector Assignment Using Output Feedback

S. Srinathkumar Langley Research Center Hampton, Virginia

NationalAeronautics and Space Administration

Scientific and Technical Information Office

1978

The problem of

SUMMARY

eigenvalue assignment in a

linear time-invariant

system

using output feedback is considered. New sufficient conditions are derived to assign an almost arbitrary set of minimum (n,m + r - 1) distinct eigenvalues where n, m, and r are the number of states, inputs, and outputs, respec- tively. These conditions precisely identify,the class of systems where such an assignment is impossible. The synthesis technique also highlights the freedom in selection of closed-loop eigenvectors under output feedback. The utility of eigenvalue/eigenvector assignment in transient response shaping is illustrated by the design of a controller for the lateral dynamics of an aircraft.

INTRODUCTION

Control system design based

on eigenvalue

or pole assignment has received

a great deal of attention in the literature.

trollable system, if state variable feedback is employed, all eigenvalues can be

assigned (ref. 1). Also it is known that for multi-input systems, the feedback

law assigning a given set of

trol laws can yield identical eigenvalues while yielding radically different

eigenvectors.

on each

fully may result in undesirable mode coupling and other poor transient behavior. For n-state feedback systems, it has been shown (ref. 2) that with m inputs, in addition to the assignment of all n eigenvalues, up to m entries in each eigenvector can be arbitrarily assigned. However, the problem of eigenvalue

assignment using output feedback instead of state feedback has not yet been com- pletely resolved. The problem of determining conditions under which all eigen- values of a system can be arbitrarily assigned to a system under output feedback has been investigated in references 3 and 4. Bounds on the number of states, in terms of number of inputs, outputs, controllability, and observability indices are established for complete pole assignability. Reference 5 shows that for a system with r outputs, if mr 2 n then the system is pole-assignable provided

the feedback gain elements are allowed to be complex numbers.

7 address the converse problem: given a controllable, observable system, how many eigenvalues can be arbitrarily assigned to the system. In general, it is concluded (ref. 7) that minimum (n,m + r - 1) eigenvalues can 'Ialmost1? always be assigned to the system using output feedback. The qualification llalmost" was introduced to cover classes of systems where such an assignment is impossible. In effect, the analysis in reference 7 does not precisely determine the condi- tions under which (m + r - 1) eigenvalues cannot be assigned to the system.

It

is

well known that

that

for

a

con-

eigenvalues is not

unique and

different con-

Since the eigenvectors determine the

failure to use

influence of

each eigenvalue

design

freedom

state variable response,

the multi-input

References 6 and

This report

considers

the problem of

determining the number

of

eigenvalues

assignable to a given system.

ment problem,

By formulating an eigenvalue/eigenvector

for

assign-

sufficient

eigenvalues are derived.

conditions required

the assignment of minimum

identify

(n,m

+ r

-

1)

These conditions precisely

the class of systems which can be assigned only d eigenvalues, where maximum

(m,r)

5

d

<

(m

+ r

-

1).

The new formulation permits

the

development of an

algorichm to assign (m + r - 1) eigenvalues. In addition, (r - 1) eigenvectors can be partially assigned with, at most, m entries in each vector arbitrarily chosen. In the event n > (m + r - I), various synthesis alternatives to stabi- lize the system are also investigated since in this case all system eigenvalues cannot be assigned. The counter example of reference 7 is used to demonstrate the utility of the new sufficient conditions in identifying systems which cannot be assigned (m + r - 1) eigenvalues. Finally, the advantage of both eigenvalue and eigenvector assignments in response shaping is illustrated by designing a controller to meet the lateral handling qualities specifications for an aircraft.

SYMBOLS

U.S.

Values are given in SI and Customary Units.

U.S.

Customary Units.

Calculations were made in

A

system matrix

 

A

E

Rnxn

n

x

n

real matrix

A

aY

lateral acceleration,

m/sec2

(ft/sec2)

B

input matrix

 

C

mea surement

matrix

C

measurement

vector

E

matrix used

in equation

(45)

 

D(k)

d,i,j,k,t

matrices

defined by

indices

equation

(B4)

vector defined in notation

(1)

in appendix B

matrix

defined

by

equation

(9)

vector

defined

by

step 3(a)

of

appendix B

matrix

defined

by

equation

(IO)

vector

defined

by

step 3(a> of

appendix B

matrices used

in equation

(23) and

derived by

substituting

equa-

tion

(6)

into equation

(7)

vector

defined

by

step 3(a) of

appendix B

(index)th

order identity matrix

j E {~(k’>

j

is an element

of

index set

A(k)

K

feedback matrix

 

Li

matrices used

in equation

(23) and

derived

by substituting equa-

tion

(6) into equation

(7)

R

feedback vector

defined by equation

(17)

M(k) ,M(k-l

,M(

matrices defined by notation

(6) in appendix B

m

number

of

inputs

max

maximum value

 

min

minimum value

N

matrix defined by

equation

(13)

n

number

of

states

P

matrix used

in equation

(46)

and defined

immediately afterward

P

roll rate, deg/sec

Q(k-1) ,Q(i)

matrices defined by equation (B3)

Q(0)

 

matrix defined

by

notation

(3) in appendix B

 

9

feedback vector

defined by

equation

(18)

r

number

of

outputs

 

S

matrix defined by

equation

(8)

S

vector

defined

immediately prior

to

equation

(20)

TO,*1

 

transformation matrices

defined where used

U

input vector

defined

 

by

equation

(2)

V

modal matrix

(matrix of

eigenvectors)

 

v(r>

 

matrix used

in equation

(12)

 

V,Vi

eigenvectors

 

W

vector

used

in equation

(6); a partition of

v

X

state

vector defined by equations

(1)

x

E

Rn

n

x

1

real vector

x

Y output vector

defined

by

equations

(1

Z(r> ,z(t)

z

matrices

containing the vectors

zi;

i

=

respectively

vector used

in equation

(6); a partition of

 

v

1

to

r

and

B sideslip,

deg

A( 1

, At

( 1

, Ar ( 1

, A( k)

set of

indices defined

in appendix B

1

to

t,

&a

aileron angular deflection,

&k

&r rudder

determinant defined

in

step

angular deflection,

deg

1

of

deg

appendix B

&Xk,&Zk

perturbed

quantities defined in

respectively

e vector

defined

by notation

(6)

step

1

and

step 4

in appendix B

of

appendix B,

Ar 9%

X,Xi,Xk

C,WI 91-12

‘k

diagonal eigenvalue matrices used respectively

in equations

eigenvalues

vectors defined by

equations

(21)

scalar defined

by notation

(4)

in appendix B

@ bank

angle,

deg

9 yaw rate,

deg/sec

(12) and

(B6),

Superscripts:

-1

A- , ,-,*

Upper

matrix inverse

transformed quantity

case letters of

the alphabet

indicate matrices;

matrix subscripts

indicate

partitioned

quantities.

Dot over a quantity denotes derivative with

respect

to

time.

Prime denotes transpose.

EIGENVALUE/EIGENVECTOR

Consider a system

linear,

time

invariant,

ASSIGNMENT FORMULATION

multivariable,

controllable,

observable

X

y

=

=

AX

cx

1

+ BU

where

trivial problem formulation,

find a

x

E

Rn,

u

E

Rm9 y

the

E

Rr,

and

assume

control

law of

form

B

m,r

and

>

1

C

are full rank; and

m,r

<

n.

for a

The problem

and

(1)

non-

is

to

u

=

Ky

(2)

in order to assign arbitrary eigenvalues for the closed-loop system. To

cate clearly the freedom available in the selection of closed-loop eigenvalues and eigenvectors under output feedback, the measurement matrix C is assumed

to be in a special canonical form:

indi-

c

c1

=

=

[Cl

[.I;

: CZ]

.I

c2 = [.1.I

c=p

I

7

11I

(3)

Here,

matrix. Appendix A details a procedure for reducing any system (C,A,B) to this special form .

t

=

r

-

1,

C1

E

Rrxt,

c

E

Rlxn-t,

and

It

denotes a

tth order

identity

The

closed-loop

system matrix

(A + BKC) after applying feedback law (2)

 

satisf ies

 

(A + BKC)vi

= Xivi

(i =

1,

2,

.

.

., n)

(4)

where

eigenvalue/eigenvector

values

freedom available in the

is

Xi

the

ith

(4)

eigenvalue and

vi

is

the

to

corresponding eigenvector.

determine the number of

and

The

assignment problem is

that

eigen-

in equation

can be arbitrarily assigned

to determine the

selection of

the associated eigenvectors.

In order

equation

(4)

see in partitioned

to

what

freedom exists form as

in the

choice

of

eigenvectors,

write

where A11,Bl E Rmxm and B1 is nonsingular. Since B is full rank, the non-

singularity of B1

can be

assured,

if

necessary,

by reordering the state vari-

ables in equations (1). Completing the multiplication of the partitioned matrices and some algebraic operations (ref. 8) permits equation (5) to be expressed as a set of constraints on the selection of eigenvectors. For clarity of presentation, these relations are detailed only for real eigenvalues. Exten- sion to complex conjugate pairs in quasi-diagonal form yielding real eigenvector

pairs is straightforward (ref. 8).

For real eigenvalues

 

[XI,_,

-

F]w

= [G

+

(6)

[AI

+ BIKCIV = XZ

(7)

where

is

the eigenvalue,

 

v'

=

[z'

:

w'];

v

is the eigenvector with

z

E

Rm,

and

 

S

= B2Bl-l

A1

=

[All

: A121

(11)

Equation (6) represents an underdetermined system of n - m equations in

unknowns.

chosen arbitrarily provided does not coincide with the spectrum of F. Examination of equation (7) reveals that at least r eigenvalues and r eigen- vectors satisfying equation (6) can be assigned to the system in equations (1) by the feedback matrix

n

be

Thus

m

eigenvector

entries corresponding to

the

z-vector

can

where A, is the diagonal matrix of r eigenvalues and

have the form T(r) = [ti : t2 :

solution to equation (12) is .guaranteed provided the eigenvalues/eigenvectors

are chosen to insure the nonsingularity of in the case of state variable feedback (C = I

assigned

Z(r)

and

V(r)

: tr] (where ti

[dr)].

It

)

all

n

are vectors).

should be noted

The

that

n. '

to the system provided the modal matrix

eigenvalues can be

eigenvectors)

(matrix of

V = [VI

: v2

:

.

.

.

:

vn]

is nonsingular.

An algorithm which

constructs

such

a nonsingular

V

is detailed in reference 8.

A pendix

B extends this

algo-

rithm to guarantee the nonsingularity of

[CV(r)f.

Notice that

by carrying out this analysis on the dual system (B1,A',C'),

it can be shown that m eigenvalues can be assigned to the system. This analy-

sis yields the following well-known result (ref. 9).

Lemma

assigned using output

1:

For system

(C,A,B),

feedback.

max(m,r)

eigenvalues can be

The analysis so far

indicates

that

only

max(m,r)

eigenvalues can be

assigned

to the system using output feedback.

However,

by sacrificing some

 

degree of

freedom

in the

selection of

the associated eigenvectors it

is possible

to extend

the number of

eigenvalues that

can be assigned to

min(n,m

+ r

-

1)

as is shown in the following section.

ALGORITHM TO ASSIGN MINIMUM (n,m + r

-

1)

EIGENVALUES

The basic approach in the

development of

this algorithm is

to 2onstruct

the

output feedback

law in equation

(2)

as

a

sum of

two

feedbacks

(K + K).

The

first feedback (E) assigns

t

eigenvalues,

and

the

second feedback assigns

 

additional min(m,n - t) eigenvalues while ensuring the protection of the t eigenvalues already assigned. The construction procedure yields a set of suf-

These condi-

ficient conditions for assigning tions also help characterize the

min(n,m + r - 1) eigenvalues. Finally, some design freedom still exists to partially assign (r - 1) eigenvectors.

min(n,m

+ r

-

1)

eigenvalues.

class of systems which cannot be assigned

Step

1:

Assign

equations

(1)

N=

t

eigenvalues and and form the matrix

corresponding eigenvectors

[.N!

N2

] =

[VI

:

v2

:

.

.

.

:

Vt]

to

the

system in

(13)

with

N.

N1

Let

E

K"

Rtxt

be

and

nonsingular.

Appendix B details

a procedure to construct

the

nonunique

feedback

(eq.

(B6))

corresponding

to

this

assign-

ment.

Then

the

closed-loop

matrix

is

&

=

A

+ BffC

(14)

Step 2:

where

Apply a coordinate

transformation

(C,$,B)

-+

(CT?,T~-I~T,,T~-~B>

The transformed

system has

the

form

where

At

E

RtXt

_I

is the diagonal matrix of eigenvalues assigned in step 1.

should be noted

that

C2

is invariant under the transformation

TI.

(16)

It

Step 3:

In order

to assign additional

eigenvalues to

the

system of

equations

(162

while protecting

is restricted to be of unity rank of the form

t eigenvalues

(A,)

already assigned,

K

= qRc

the

with

second feedback

q

E

Rm,

R E

(K)

Rr,

and

is chosen so that

 

R'P1

:

c2]

=

[o

:

c]

(17)

Now

q

must

be

chosen so that

 

is assigned

Since

min(m,n

(i22 ,g2)

-

t)

eigenvalues.

is controllable,

the

following result holds:

 

/.4

Theorem 1:

The single output subsystem (c,A22,B2)

can be

~

assigned

min(m,n

-

t)

eigenvalues if

and

only if

(a) (c,A22)

is observable and (b)

62

is

full

rank.

Theorem

1 follows directly from lemma

1.

Further,

conditions

(a) and

(b)

restrict

the admissible set of

eigenvalue/eigenvector assignments in step

1.

For

conceptual convenience,

these parametric restrictions are formulated in

terms of

controllability conditions of

a

variable representation.

fictitious dynamic

system in the

state

the

After matr;x

submatrix

A22

operations in equations can be written

as

(14) and

(16) have been performed,

(20) can now be

looked upon

Equation feedback to the

dynamic

system

1

 
 

i =

Ai25

+ Ai21-11 + c'1-12

n

= $5

+

Bi1-11

with

6

E

Rn-t,

q

E

Rm,

1-11 E

Rt,

1-12 E

N2

= -N2N1'1.

 

as a system matrix derived

by applying

(211

R',

1-11

=

fi&,

1-12 = slq, and

From equations

(21) it

can be

shown that

are equivalent to the following

conditions:

conditions

(c)

&

=

B2

(a)_and (b) of

+ N2B1

and

is

rank;

(d)

[[A22

+ A{~G;],C~

is controllable.

Conditions

(c) and

(d)

theorem

full

clearly

1

indicate the restriction on selection of the eigenvector parameters in step 1. This yields the following sufficient conditions:

Theorem 2: The system (C,A,B) can be assigned min(n,m + r - 1) eigenvalues arbitrarily close to the desired set if the first t eigenvalues and eigenvectors in step 1 are chosen so that

I

N1

is

nonsingular

 
 

nl

I1

62

=

B2

+ N2B1

and

is

full rank

I11

{[A;2

+

Ai&]

,cl}

is

controllable

"Arbitrarily

closer1 in

theorem 2 indicates

needed in

eigenvalue

specifications may be

the

that

slight perturbations

following

situations:

in

(i) Assigned

eigenvalues coincide with

the

spectrum of

F

(appendix B).

 

(ii) An

exact

combination of

eigenvalue/eigenvector

specifications

in

step

1

may

not

yield

a nonsingular

N1

(appendix B).

 

(iii) Coincident spectrum situation similar

to

situation

(i) exists

for

the

subsystem eigenvalue assignment

of

theorem 1.

Condition I is required to guarantee the existence of the transformation TI in step 2. This condition can be explicitly included in the synthesis pro- cedure as detailed in appendix B. Condition I11 is obtained from the property that for system (eqs. (2111, the class of feedback from input 1-11 should be restricted so that the controllability of the feedback system with respect to

the input 1-12 is preserved. Conditions I1 and I11 yield nonlinear_ algebraic

and thus,

constraints for the elements of the eigenvector parameter matrix

in general,

However, example 1 of the section entitled llNumerical Examples1! shows how these conditions can be explicitly checked.

N2

can only be used

as test

conditions for

each assignment

in step

1.

Step 4:

If theorem 2 holgs,

Amin(m,n - t)

eigenvalues can be assigned

to

the

sin-

gle output system (c,A22,B2) by the feedback

q

derived

from lemma

1.

Step 5:

The composite

feedback law

 
 

u

=

(i? + i)y

 

(22)

assigns

min(n,m

+ r

-

1)

eigenvalues to

the

system.

 

ASSIGNING

n

EIGENVALUES

The development

so far has revealed

that

for

systems where

n

>

(m

+ r

-

l),

all system eigenvalues cannot usually be assigned.

tions

values of

However,

by using equa-

(6) and

the

(7),

it is possible to derive conditions for assigning all eigen-

system as follows.

If

the

eigenvalues to be assigned

in

F (if necessary,

by a perturbation

are noncoincident with

specification), then

the

spectrum of

equation

(6) can be

explicitly solved

for

the

w-vector

and substituted

into equation

(7).

This

sub-

stitution yields a

set of

homogeneous

equations of

the form

 
 

[Hi

+ B

 

KL-

2-i

=

0

(i = 1, 2,

 

.

., n)

 

(23)

 

1

11

.

 

where

Hi

and

Li

are readily

derived.

Then equation

(23) has a nontrivial

solution

if

and

only

if

 
 

rank

[Hi

+ BlKLi]

< m

for all

i

(24)

A

set

of

n

nonlinear

equations in

the

m,r

parameters of

the gain

matrix

K

can be

derived by

setting the appropriate

determinant in equation

(24)

equal to zero.

tence of

However,

the

solution.

no general conclusions can

be drawn regarding

the exis-

to

assign t eigenvalues as in equation (14) and to attempt to assign the remain- ing (n - t) eigenvalues approximately to the subsystem in equation (18) using the condition in equation (24). In this case, a set of (n - t) linear equations

in m unknowns results, and a least-squares solution can be obtained. The foregoing discussions assume that r 2 m if necessary by considering the dual system.

An alternative approach not

involving solution

of

nonlinear

equations is

Consider

-

A=

B=

the

NUMERICAL

EXAMPLES

Example

1

system described in reference 7 where

1

0

0

0

0-

0

0

1-

-

00

01OL

O

0,

 

1

0

-

c=

 

0

0

O10O1-

1

Reduce the system

variables as

ing the coordinate transformation (C,A,B) + (ET0 ,To-lZTo ,To-IG) where

(25)

to the

special form of

-

make

equations

(1) by ordering state

and by apply-

(XJ,X~,X~,X~)to

sa- nonsingular

(appendix A)

::::0001 0010

(26)

to yield

A=

B=

001

000

000

000

c=

From equation

1

0

(6)

00

11

can be

O11 ,

shown that

the

closed-loop

eigenvectors

satisfy

where

the

eigenvalue

is

X

From theorem 2,

of

choices of

generalitx.

N2

/

condition I,

Then,

since

-

N2

Bi

=

=

0 0

and

the

eigenvector

is

n1

#

0;

therefore,

choose

n1

=

1

without

loss

( 42

-9

-n4) I.

Condition I1 is met

for

all

0.

Condition

I11 implies

that

11-12

l!

-n3

-n4

0 0

should be controllable.

of

For

the

controllability matrix in equation

in equation

(30) to

full

rank,

the eigenvector parameters

(28) must satisfy

be

n2n3

-

n4

+ n32

+ n3n4

#

o

(311

By direct substitution

sible selections in equation (28).

(m + r - 1) = 3 eigenvalues. Indeed, only two eigenvalues

system cannot

equation (31) is

seen to be violated by all admis-

the

Thus,

be assigned can be assigned to

this

precisely lead

system (ref.

to

7).

this

However,

conclusion.

the

previous analyses

Example 2

Consider the system described by

(ref. 7)

1

0

0

0_1

:1l -

0

1

0OI.

It is required to assign eigenvalues close to -1,

-2,

(refs.

and -5.

3 and

7) do

not

The system (32)

can

be

reduced

to

the form of

equations

(1)

by following the procedure in appen-

dix

A

as

Step

1:

Step 2:

Measurement matrix

is in the desired form with

Apply coordinate

transformation

ea

nonsingular.

(C,A,B) + ~CT~,T~-~~T~,T~-~B~

where

100

011

001

yielding

A=

B=

11

01

00 "11

c=

1

0

"1/

11

O

The system (34)

loop eigenvector constraints

respective matrices

B1 = [;

is in the required

in

0

-1

canonical form of

equations

in equations

to

(34) can be derived

equations (8)

(10) as

(I).

The closed-

by

identifying the

s

=

(1

F=O

-1)

G

=

(0

-1)

and

equation

(6) yields

Equation

X

#

0.

xw

=

[A

:

-(1

+ A)] 1.": .1

(35) indicates

that

Lz2

z1

1

and

>

can be

Now,

applying the algorithm to assign

(m

+

(35)

rbitrarily

r

chosen,

provided

- 1) eigenvalues yields

the

following synthesis sequence.

 

Step

1:

Assign

A

=

-1.

From appendix B,

case 11,

At(1)

=

{l},

and

this

implies

z1

is

#

0

in equation

(35).

(Condition I,

theorem 2.)

One acceptable assignment

N =

(1

1

1)'

(36)

and a nonunique feedback gain corresponding to this assignment from equa- tion (B6) is

The closed-loop system matrix of equation (14) is

Step 2:

Transform the system to the canonical form of equations (16) using

TI

to yield

1.0 -11 1.1 0 1.0
1.0
-11
1.1
0
1.0
 

-1

.

0

OI-

.

.

.

.

.

.

.

0.-1

 

0

0.

 

0

0

-

1

0

jL

 

-1

-1

-

0

1

e=[;

 

;

;j

From equations (21) we can form the dynamic system with Since feedback from I-rl does
From equations
(21)
we can
form the
dynamic system
with
Since feedback from
I-rl
does not
affect
the
controllability of
system (41)
with respect to ~2,the subsystem is pole assignable from condition I11 of
theorem 2.
Further,
the
assignment
in step
1 satisfies
condition
I1 since
82
is nonsingular.
Thus theorem 2 holds.
Step 3:
Choose
R'
=
(-2
1).
This choi2e protects
X =
-1
assigned in step
1.
It
now remains
to
choose
q
so that
K
= qR'
assigns
the
eigenvalues
X2
=
-2
and
A3
=
-5.
Step 4:
Assign
A2
=
-2,
A3 = -5
to
the
single output subsystem in equation
(18)
where
c
=
E1
13
16

with

feedback

Step 5:

The feedback law of

J

the

form of

equation

(22) is

given by

and assigns the eigenvalues -1,

-2,

and

-5

to

the

system (eq.

(32)).

The advantage of

AIRCRAFT

LATERAL,

combined

control of

CONTROL

DESIGN

closed-loop

eigenvalues and eigenvec-

tors using state variable feedback has been investigated in reference IO. The utility of the output feedback extensions developed in this report will now be illustrated through the design of a lateral controller for an aircraft.

The linear perturbation model

be modeled

j:

as

=

Ax

+

Bu

y = Ex + Du

1

for

the

lateral motions of

an aircraft can

(45)

where

bank angle 0, respectively. The control vector of aileron 6, and rudder

6, angular deflections is u.

tion ay constitute the output vector y. All angles are in degrees, rates in

x

is

the

state vector of roll rate

Roll

rate

p,

p,

yaw rate

yaw rate

9,

sideslip

6,

and

Q, and

lateral

accelera-

deg/sec,

and acceleration

in m/sec2

(ft/sec2).

The respective matrices

in equations

(45)

for a

fighter aircraft

at an alti-

tude of 6096 m (20 000 ft), a Mach number of 0.67, and an angle of attack of

3.45O are given by

where

r-3*79

A=-I I 0.06

-0.14

25

0.04

-0.36

-1

0.06

-52

4.24

-0.27

0

0

0

0.05

0

0.01

0

L-0.13

0.05

0

0 0

0

1 0

0

-0.06

-3.42

0

the

11.03

elements of

Since the output

the matr-ces

are approx-aated

to

two significant dig ts.

'vector

in equations

(45)

is derived as a linear combina-

tion of both state variables and control inputs, the closed-loop system after

applying feedback u = K*y takes the form

 

x

=

(si

+

BKC)X + GPu

 
 

(46)

 

y

=

(e + ~K(?)x

+ DPu

 

where

P

=

[Im

- K*D]-'

 

and

K

=

PK*

is the equivalent output feedback matrix

obtained

 

by

setting

=

0.

Thus,

the algorithms developed earlier for systems

with

5

=

0

are applicable to systems of

the

form of

equations

(451,

provided

P

exists.

 

Then the feedback gain

K*

= K[Ik

+ fiK1-l

K*

is

computed

by using the relation

(47)

Direct matrix manipulations show that exists.

the

inverse in equation

(47) exists

if

P

The handling qualities specifications (ref. 11) imply that the lateral air- craft dynamics should be composed of two weakly coupled subsystems. Roll rate and bank angle constitute the first subsystem and display predominantly the roll subsidence and spiral modes. The second subsystem is characterized by a well- damped Dutch roll mode defining the yaw rate and sideslip motions. These speci- fications can now be formulated as an eigenvalue/eigenvector assignment problem.

Table I summarizes the modal

characteristics of

the

free aircraft.

From

the table it is seen that the Dutch roll mode is very lightly damped and appears

dominantly in the response of

dominant entries in the corresponding eigenvector pair. Thus, the eigenvalue/ eigenvector modification requires that the closed-loop system have the modes

and mode-variable associations of table 11. (In tables I to

the roll variables

p and 9 as evidenced by

111,

j = n.

TABLE 1.- MODAL CHARACTERISTICS OF FREE AIRCRAFT

Eigenvector

components

P

iJ

B

9

Eigenvalue of

-

-3.70

(Roll subsidence)

-0.35 5 j2.66 (Dutch roll)

. I

-0.03

.

(Spiral1

-0.964

-0.403

0.829

-0.032

-.041

-.096

-.I31

i

-.002

-069

-.034

.002

.261

.998

1

Mode

Roll subsidence

TABLE 11.-

DESIRED MODAL SPECIFICATIONS

Eigenvalue

-6

Dutch

roll

-1.0

-+ j0.2

Spiral

-0.01

The output feedback analysis in the main text

shows that

all

system eigen-

values can be assigned since n = m + r - 1 and only two eigenvectors (t) can be assigned with at most two (m) entries in each vector arbitrarily chosen. The eigenvector freedom available was used to control the structure of the eigenvec- tor pair corresponding to the Dutch roll mode to effect the desired yaw rate and sideslip dominance. The modal coupling matrices D(k) (eq. (B4)) aid in the selection of the Dutch roll mode to yield the appropriate eigenvector forms. The four eigenvector entries that were freely chosen corresponded to the roll rate and yaw rate components of the real eigenvector pair associated with the Dutch roll mode. Since gain magnitude constraints cannot be explicitly included

into the synthesis algorithm, the design parameters have to be iteratively modi- fied to meet gain limit requirements. After some design iterations, a compro- mise design yielded the modal characteristics summarized in table 111. For

example, Zt was noted

improve $ and

that

the Dutch roll mode damping could not

without violating

feedback

be reduced

(to

were

B

responses)

gain limits which

set at unity for this analysis.

(eq. (47)) as

The design yielded

a feedback gain matrix

K*

TABLE

1-0.16

0.19

-0.6

1

111.- MODAL

CHARACTERISTICS

OF FEEDBACK AUGMENTED AIRCRAFT

the

7 Eigenvalue of

-

Eigenvector

components

-6

(Roll subsidence)

+ j0.20

(Dutcg roll)

-1.0

0.986

0.01

0.02

-0.01

1

(Spiral)

-0.013