Sei sulla pagina 1di 3

A Controllability Criterion ror a Class BT(t) =(9)b2T

0r Linear Systems Bt[ bm7 _

A. R. STUBBERUD and B(t)BT(t) can be written


MEMBER IEEE B(t)BT(t)= [bib1T+b2b2 T+, ..., brbmT]

=
mbibT (10)
Summary: A linear system is completely vector, A(t) is an X n matrix with ana-
n
controllable if for any initial state a control lytic elements, and B(t) is an nXm
can be found to take the state of the system . . ' . . Using equation 10, condition 1 is
to the origin of the state space. Presetnt matrix with analytic elements. Uslng U
criteria for testing linear systems for the notation of equation 1, the following m
controllability require knowing the weight- definition of coiltrollability is given: Zftt0 zT (to, t)b0biT+T(to, t)zdt
ing function matrix of the linear system. Definition 1:1 A state yo is said to be i=1
In this paper a new criterion is presented m
which is applicable to a broad class of linear controllable at time to if there exists a con- yftt [zTp(to, t)bi] 2dt>0 (11)
systems defined by linear n-dimensional trol function x'(t) depending on yo and to i=
ordinary vector-matrix differential equa- and defined over some finite closed inter-
tions; for this criterion only the coeflicient val (to, tl) such that y(x', t1 Yo, t) =0 If for all z#0. Note that this condition
matrices of the differential equation need this is true for every state yo, the system implies that the ternms
to be known.
is completely controllable at time to; if zTq(t0, t)bi i=1, 2, ...,
this is trie for every to, the system is
HE CONCEPT of controllability of completely controllable. are not all identically zero over the in-
| lincar systems has been developed It can be shown that a necessary and terval [to, ti]. With the preliminaries the
by R. E. Kalmanli- and others.4-6 Con- sufficient condition for a linear system following theorem is now developed.
trollability is that property of a system defired by equation 1 to be completely Theorem 1: A necessary and sufficient
which allows its state to be transferred controllable at time to is that the matrix condition that a svstem described by
between two points by means of some z equation 1, and subjcct to the condition of
suitable control input; the concept is ll(to, tl)= ft0 0(to, t)B(t)BT(t)kT(to, t)dt (2) analyticity of the elements of A(t) and
valuable since it indicates the existence be positive definite for some t1>t0 where B(t), be completely controllable at time
of a control which can make this transfer. BT(t) and T(t0, t) are the transposes of to is that the nX (mn) matrix
An obvious problem, however, is that of B(t) and 0(t0, t), respectively. The .,= [ 1T2, n] (12)
finding a suiitable criterion for determin- matrix 4(t, to) is defined as the weighting
ing the controllability of a particular sys- function matrix of equation 1; i.e., where n=B(t)
tem. General controllabilitv criteria
have been developed,'2,4 but these -
are ~~~~do(t, to) dAtkt t)() -ii=-At)~1(3
A(t) rj+
dt (13)
=
A(t)o(t, to) (3) Yi -

usually difficult to apply by computation. dt


In this paper, a criterion is developed 0(to, to) = I (the identity matrix) i = 1, 2, n -1
wliich is applicable to a large class of It can be shown that have rank n for t>to.
linear systems and, in many cases, rela-
tively simple to apply by computation. o (t, to0)(to, t) = (4) Proof of Sufficiency: Assume that y
and,Cconsequently,othatlabttyt)hsatisfiesnthasis not completely controllable at time
an,
rank n but that the system of equation
cy

Condition of Controllability the equation to. Then for all t>to, there is at least
A large class of linear systems is defined d4(t, t)= -(to, t)A(t) (5) one z#0 such that
by the nth order vector-matrix equation dt zT+(t0, t)b0_0 (14)
of theform 45(to, to) =I z0t,tb=O(4

dy In order for the matrix W1(to, t1) in i= 1, 2. m


dt-=A(t)y+B(t)x to<t<+ o (1) equation 2 to be positive definite at time If equation 14 is differentiated n-I times

y(to) = yo t0, the quadratic form with respect to t, the following set of mn

zT 111(to, tl)z (6) equations is formed:


where y is the n-dimensional state ti s
vector, x is the m-dimensional control where z is a constant n-vector, must be zT? [4(to, t)b] =0 (15)
greater than zero for all z5# 0. Sub- dt
A paper recommended by the IEEE Feedback stituting equation 2 into equation 6, this j=0, 1, n-i i= 1, 2, .. ., m
Control Systems Committee and approved by condition can be written
the IEEE Technical Operations Committee for Dfn
presentation at the IEEE Western Electronics fr tlTT TDfl
Show and Conference, San Francisco, Calif., J oz (ts, t)B(t)B (t)+ (to, t)zdt>0 (7)
August 20-23, 0963. Manuscript submitted June t(
16)
b,=
24, 1963: made available for printing April 30, 0964. for all z #0.=
A. R. STUBBERUD is with the University of Cali- Let the matrix B(t) be partitioned into and examine the vector
fornia, Los Angeles, Calif. clms,..
The work reported in this paper was supported in ,*?dd2
part by grant AFOSR 62-68 under U. S. Govern- Bt) - [b, b,. , b,~ (8 dq.to d)t]= [(otn]17
ment Contract No. AF 33(657)-7154. Repro- ()[1 ,.,m 8 td
duction in part or in whole is permitted for any
purposes of the United States Government. where bs= b0(t). Then BT(t) is Forj= 1,
NOVEMBER 1964 Stubberud Controllability Criterion for Linear Systems 41 1

Authorized licensd use limted to: IE Xplore. Downlade on May 13,20 at 1:57 UTC from IE Xplore. Restricon aply.
d d dy.i t) y are analytic; therefore, if they are concepts of nornal and proper linear
d[(to, t)Tn ] =dt [4(to t)]TnB+f(to, t) dt zero for an interval of time of positive systems. In the notation of this paper,
(18) length, they are zero for all tinie to.t<+ the following definitions can be made.
00. That is, if the row vectors of O(to, t)y Definition 2: A system is called proper
which, with equation 5, can be written are linearly dependent over a positive if
interval, they are linearly dependent for T B - T 3
-(t1,
t)A(t)TnZ±f(ti, t) dt
dt
all time to<t<+ o. Conversely, if they
are linearly independent over a positive on an interval of positive length implies
d=dt -A(t)iA/ (19) interval, they are linearly independent for z = 0.
t(
all time to.t<+. Definition 3: A system is called
If If a set of mnXn matrices are formed normal if, for each i,
dy. -A(t)Ynt (0 'Y(T1Tf2'..,i) i=li, 2, ..., m zT,k(to, t)b (t)=O (32)
Yn-1t dt
(20) then equation 25 can be written on an interval of positive length implies
then zTO(t0, t)(-y1.y2, Z= 0.
thenZT[tOt)( ty10t,,TMt)-2, It follows that, for systems described
-d[(to, t)yn2]=-dt [4(to, t)dniZ
dt2
ZT[O(t0, t)y1¢(to, t)ym.- (to, t)0ym](25A)
=0
by equation 1 with the given constraints
on the elements of A(t) and B(t), the con-
X ( . cepts of proper and completely controll-
If there is a constant solution vector
= o(t, t) dYnl_ -A(t)T,i1 z
able systems are equivalent. The en-
dt satisfving this> equation,' then it satisfies ~~~~~~~~terionof Theorem I for testing for com-
= 0(to, t)Yn-2t (21) zT (to, t) -O= (28) plete controllability becomes a criterion
In general, therefore, for arbitrary i. The determiniant of for testing for proper systems by sub-

Of(th
t
*y
)e
is the Wronskian determinant
spletely
stituting "proper" for the phrase "com-
controllable at time to'' in the
-[ (to, t)Yn7 = O(to, OYnJ (22)
theorem.~~~fth
lmet o elem
hevcorkto )b()
A solution of this equation, if it exists, is therem.
~~~~scribed by equationtesting
Equation 15 can then be written a constant vector, which is z. Solution A criterion for a system de-
' . 1 for normality iS
ZTO(to, t)n = 0(23) of equation 25A must be a multiple of z
of Necessity:
Proof Proofof Ncessty: Assume
ssum tha the gieasflo.
that the
i =1, 2,., m TTheorem 22: A A necessar-v and sufficient
j=O, 1,..., n-1 system of equation 1 is completelv con-
condition that a svstem described by
trollable at time to but that the rank ofI equation 1 and
Now form a set of matrices from the tre matr timetb tha T therk subject to the differenti-
column vectors y-Ii as follows: t m . ability conditions which follow equation I
is a solution zF#O to equation 2.A and,
'Yn-j=(TYn_jyn-yj Ynj) (24) consequently, there is a z#O such that
j=O,l, . n-I ., zT 0(to, t)bi=0 (29) i=(YIiy2i, Yn')i =1, 2, .. ., m (33)
then equation 23 can be written in vector for all i, and W(to, t1) is not positive defi- where
matrix form as nite. This contradicts the assumption
z
zT~(t0, T)(yi.y~ . =
T0(tO, t)(_12, (25)
Yn = OT (25)
that the system is completely controllable Yn= bi(t)
at time to, and necessity is proved. dy+l'
where 0 is a n-dimensional zero vector. For the case when A(t) and B(t) are Yi dt -A(t)yj+11 (34)
The matrix 0(to, t) is a fundamental constant matrices, the matrix described
matrix and therefore, has rank n. By by equation 12 becomes j= 1, 2, ..., n-i
hypothesis, the matrix a= [(-A)n 1B, ...,-AB B] (30) each has rank n.
.-IT2,7n) (26) Thus, a necessary and sufficient condition The proof of this follows directly from
has rank i; therefore, the matrix
for a time-invariant linear system de- combining definition 3 and equation 23:
scribed by equation 1 to be completely zT (to, t)bi(t) = zTO(to,
t)yn_i = 0 (35)
(to, t)(yTIy2. T,O) (27) controllable is that the matrix in equation
33 have rank n. Kalman' proved this j=O,1, n-i
has rank n. The only solution of equa- result. *
tion 25 then, is z o, which contradictsr x form,
the assumption that z#O, and thus suffi- NONANALYTIC SYSTEMS zTq(tO, t)>1 =O (36)
ciency is proved. If A(t) and B(t) are not analvtic but
Before examining the question of neces- have n -2 and n -1 continuous deriva- The equation has a solution z n 3 if, and
sity of the theorem, the following observa- tives respectively then this criterion is only if, the rank of at is less than n, thus
tions will be made. Since the matrix sufficient but not necessary (this was proving the theorem.
-AT(t) in equation 5 has analytic ele- pointed out to the author by A. Chang of Again, if A(t) and B(t) are nonanalytic
ments, the matrix (to, t) has analytic the University of California Berkeley but have i -2 and - continuousderiv-
elements.7 Also, since derivatives, sums, Calif.). atives, then these criteria for normal and
and products of analytic functions are proper systems are sufficient hut not
analytic, the matrix ry is composed of ana- Relationship to Normal and Proper necessary.
lytic elements only. Apparently, then, Systems Example: As an example of the use
the product matrix k(to, t) ,y also has only of Theorem 1, consider the time-variable
analytic elements. Now, for a constant LaSalle, in his treatment of time third-order vector-matrix differential
vector z the elements of the vector zT4(to, optimal control systems,4 discusses the equation
412 St1ubberud-Controllability Criterion for Linear Systems NOVEMnER 1964

Authorized licensd use limted to: IE Xplore. Downlade on May 13,20 at 1:57 UTC from IE Xplore. Restricon aply.
Yo ~-(1 +ee- t) 1 0 rYO_ The controllability criterion states that References
Y1 = ( 1 + 3e- t ) O -1 Y, + the 3 X 3 matrix formed of these vectors
3e-'
_~2 0 0_ -.y2_ mulst be nonsingular; i.e., the three 1. CONTRIBUTIONS TO THE THEORY OPTIMAL.
CONTROL, R. E. Kalman. Proceedintgs,OPConference
O vectors must be linearly independent. on Ordinary Differential Equations, Mexico City,
< I. I . . Mexico, in Boletin de la Sociedad .fatematica
Ui x (37) However, it is apparent by inspection that Mexican:, 1960, p. H,2
Le j 2. CONTROLLABILITY OF LINEAR DYNAMIC SYS-
'(1 = - (39) TEMS, R. E. Kalman, Y. C. Ho, Kr. S. Norenda.
~thegeneral
Using ~~ ~~~ 7
. _. Contributionzs Differeto
Using crtrindecibdb
th
criterion described by to tial Equations, Interscience
equation 2 it would be necessary to deter- Thus, the system shown by equation 37 iS Publishers, Inc., New York, N. Y., vol. 1, Aug.
mat**rix of ts
mine the weighting function not fully controllable for any time to. 1962, pp. 189-213.
3. MATHEMATICAL DESCRIPTION OF LINEAR
equation. To apply the criterion of this DYNAMICAL SYSTEMS, R. E. Kalman. Technical
paper it is necessary to form the vectors Conclusion Report No.Baltimore,
Studies, 62-18, Research
Md., Nov.Institute
1962. for Advanced
(1 as defined in equation 12. These are A criterion for testing the complete 4. THE TIME OPTIMAL CONTROL PROBLEM, J.
controllability of a general class of linear P. LaSalle. Contibutioins
~~~~~~~~~~~~~~~~~~~~~~~linear
Oscillations,
to theUniversity
Princeton
Theory of Non-
Press,
0
= systems has been developed. This cri- Princeton, N. J., vol. 5, 1960.
e-t terion may also be used, for the same
class of systems, to determine if the
5. WHAT CONSTITUTES A CONTROLLABLE SYSTEM?
(Y.(Correspondence),
C. Ho. Transactions on Automatic Control
to
detrmine f the Institute of Radio Engineers,
__ 1 _ system is normal or proper. The utility vol. AC-7, Apr. 1962, p. 76.
Y2 = e of this technique in comparison with the 6. CONTROLLABILITY AND OBSERVABILITY IN
-e . . MULTIVARIABLE SYSTEMS, E. G. Gilbert. Journal
standard tests lies in the fact that the on Control, Series A, vol. 1, No. 2, 1963, pp. 128-51.
-1 weighting function matrix of the system 7. THEORY OF ORDINARY DIFFERENTIAL EQUA-
1= +e t need not be known, . ' only.t]
the coefficient TIONS (book), E. A. Coddington, N. Levinson.
Inc., New York,
N. Y., 1950, p. 90. Company,
McGraw-Hill Book
_-2e t_ (38) matrices of the differential equation.

.
athematical .. ntion accounts for the fact that a system
M
at ematical Interre ations ip Between etw een may be repaired back into an operable
condition once it has failed.
sel i a li i ty a n o Ivi a i n t a i n a il i ty I n 5. Active constant redundancy: Each
redundant unit is active from time zero,

various System oels and the individual failure rates do not


change.
Visual Aids
S. J. HORNYAK A state-space diagram, which illus-
MEMBER IEEE
trates the various states in which a system
may lie and the means of transition (if
Summary: This paper defines and sum- THE FOLLOWING definitions apply any) between states either by failure or
marizes several concepts and tools which...
have seendeverlonepts intool
marebeen developed to aidand
in yswh ofr
* the analysis to this paper: repair, is the clearest means of presenting
electronic systems from the part level 1. Reliability: The capability or the probabilitv equations of a system.
through the system level. The problems .
.. of a unit. or system performing
probability Because th
sierd small increments
prbblt time are
of aofdobl con-
t....1ransi
explored are those for which the parameters, sidered, the probablity Of a double transi-
failure rates, and repair rates mayv be its intended function for a stated period tion can be assumed as zero. The space
assumed constant; i.e., the exponential dis- of time without failure, when operating diagram for a 2-unit concomitant (simul-
tribution may be assumed. It also is under the conditions for which it was de-
assumed that all parts, components, etc., are signed taneously operatiiig) system is illustrated
operable at an arbitrary reference time zero. gnd in Fig. 1. The possible states are
Moreover, a short cut for determining the 2. Nlonmaintained system reliability: 1. Units A and B are operable and operate
mean of the type of equations discussed in
this paper is presented,
The unattended unit or svstem reliability,
the (state 3).
S
defining limiting availability is included,s of. .
.. . the.direct. method i.e.,
e the reliability
el oof a svstem on which no 2. Unit A has failed and B is operating
and the method of determining the mean of maintenance (other than preventive) is (state 2).
the reliabilities from their Laplace transform performed until after the svstem has
is demonstrated, along with checks on the 1 . . failed.
. ' ~~~~~~~~~~3.
Unt A is operating and B has failed
Unit Ath
availability equation matrix of an active (state 1).
system. 3. Milaintained system reliability: The 4. Both A and B have failed: a system
reliability of an attended or maintained failure (state 0).
system; in part, the probability that if Fg osntdsrb opeeygn
Paper 63-625, recommended by the AJEE Systems one member of a redundant pair has failed; era prbe beas teim torpr
Science Committee and approved by the AIEE. . ' erlpoembcuetetetoeai
Technical Operations Committee for presentation the other member will not fail (system fo ahsaei osdrdeul
N. Y., January 27-February 1, i963. Manuscript flue) before thefrtebrreaed hwv,echuiissigddfeen
submitted April 5, i963; made available for print- Aviairy
. h rcinlaon aiuerts h iga losrpi
ing July 8, 1964. 4 vllblt:Tercinlmut fiuerts h iga losrpi
S. J. HORNYAK was with the Martin-MVarietta of time the system will be available for from each state; the appropriate repair
Company, Baltimore, Md., when this paper was operation even though system failures rates are set equal to zero, depending upon
completed, and iS now with the Bunker-Ramo haeocreorwlocu;tideni wihoeofheheesuinsstob
Corporation, Canoga Park, Calif.haeocreorwlocu;tideii whconofheheesuinsstob
NOVEMBER 1964 Hornyak-Reliability and M7laintainability in System Mlodels 413

Authorized licensd use limted to: IE Xplore. Downlade on May 13,20 at 1:57 UTC from IE Xplore. Restricon aply.

Potrebbero piacerti anche