Professional Documents
Culture Documents
00
Printed in Great Britain. Pergamon Press plc
(~ 1991 International Federation of Automatic Control
Technical Communique
et aL (1985), one of the following holds when 3., (or NI) < 0:
(i) hi(0 ) < 0, h2(0 ) < 0, det (H(0)) < 0;
(ii) hi(0 ) > 0, h2(0 ) > 0, det ( n ( 0 ) ) < 0;
(iii) either one of hi(0) and he(0) is negative and
det (H(0)) > 0
FIG. 1. Decentralized feedback control scheme. where hi(s ) = gii(s)ci(s) for i = 1, 2 and H(s) =
G(s) diag [cc(s)]i=l.2.
Since det ( H ( 0 ) ) > 0 is only a necessary condition for
Under Assumption 1, the decentralized stability conditions decentralized stability for 2 x 2 and larger systems while
are (Manousiouthakis, 1989) hi(O) > 0 is a necessary and sufficient one for the stability of
the individual loops, (la) follows from (1), ( l b ) from (ii),
(h2(s)gzl(s) = (121(s)gle(s) (8)
(lc), (ld) from (iii). []
ghl(S)g122(s) - 012(s)~21(s) = - q l z ( s ) = -qzl(S) (9) The next proposition follows directly from Lemma 1.
g~ds) ge,(s) "
The decentralized IMC controllers in this note and the Proposition 2. If Assumption 1 holds and the stability of the
decentralized feedback controllers in Manousiouthakis decentralized control system is required, then one of the
(1989) are related by the following equality following conditions holds:
1. Zii (or N I ) < 0 , one of the loops 1 and 2 has to be
Q ( s ) ( 1 - Gm(s)Q(s))-' = ( 1 - O_(s)G(s)) 1(2(s). (10) unstable;
2. ;~ii (or N I ) > 0 , both loops 1 and 2 have to be unstable;
It is noted that the decentralized stability constraints on ~0(s) 3. ).~ (or NI) > 0, both loops 1 and 2 have to be stable.
[(8), (9)] will render the right hand side of (10) a diagonal It is noted that (3) of Proposition 2 was discussed in
matrix. Grosdidier et al. (1985) and Niederlinski (1971), while (1),
After some algebra (10) yields (2) of Proposition 2 remain open. This is not surprising. The
assumption that the individual loop is stabilized in Grosdidier
q,(s) qll(S) - -tl2'(s) (11) et al. (1985) and Niederlinski (1971) precludes the discussion
1-gl,2(s)gzt(s) glz2(s)g2,(s) of (1), (2) in Proposition 2. The next section gives the main
qz(s) - qee(S) -qlz(s) (12) result of this note.
1 - (]21(s)gl2(S) (h,(s)g12(s)"
4. Main result
Therefore, from (7), (8), (9), (11), (12)
Theorem 1. Let Assumption 1 hold. The necessary and
1 sufficient condition for decentralized stability is either one of
a(s) = (13)
1 - ¢]21(S)gle(S) the following conditions:
(1) If ;ti~ (or N I ) < 0, one of the two loops has only one
den ((121(s)gle(s)) unstable closed-loop pole, while the other loop is stable;
= den ((121(s)g12(s)) - n u m ((lel(s)g12(s)) (14) (2) If ).ii (or NI) > 0, both loops have to be stable.
where n u m ( . ) and d e n ( . ) denote the numerator and Proof. The proof is by contradiction.
denominator of (.) respectively. Since both ~21(s) and glz(S) (1) If ).ii (or N I ) < 0 , one of the two loops has to be
are stable rational transfer functions, it can be concluded that unstable from (1) of Proposition 2. Suppose that one can
A(s) does not contain any RHP-zeros and the proof is assign more than one unstable pole, say two unstable poles to
complete. [] loop 1 (the same reasoning applies in a straightforward
The next lemma relates the stability conditions of a 2 × 2 fashion to more than two unstable poles cases):
decentralized control system to R G A and NI.
Lemma 1. If Assumption 1 holds and nl(s)n2(s) Re {Pi} > 0 , i = 1, 2
qt(s) = (s - pl)(S - p2)dt(s)
1. 3., (or N I ) < 0 , then one of the following conditions
holds: qz(s) to be any stable, proper, rational transfer function
(a) the loops 1 and 2 and the system are unstable; where Re {.} denotes the real part of (.}, dl(s ) is a
(b) the loops 1 and 2 are stable, but the system is polynomial with only LHP-zeros, and nl(s ), nz(s) are the
unstable; polynomials such that ql(s) and q~(s), q~(s) defined below
(c) only one of the loops 1 and 2 is stable, but the system are proper, ql(s), qz(s) are designed such that A(s) of (7)
is unstable; satisfies Proposition 1. Nevertheless the following design also
(d) only one of the loops 1 and 2 is stable, but the system stabilizes the plant:
is stable. n,(s)
q~(s) =
(s - p O d , ( s )
2. 3., (or N I ) > 0, then one of the following conditions
holds: (la), (lb), (lc) and q~(s) = ~ - qz(s)
(a) the loops 1 and 2 and the system are stable;
(b) the loops 1 and 2 are unstable, but the system is because ql(s)qe(s) = q~(s)q~(s) and hence A(s) = A'(s).
stable. But then this contradicts (1) of Proposition 2. Therefore
Proof. We only show the proof of part 1, since the same one has to destabilize any one loop with only one unstable
reasoning follows directly to prove part 2. From Grosdidier closed-loop pole in order to stabilize the plant.
(2) If 3., (or N I ) > 0, one possibility is that the two loops
have to be unstable from (2) of Proposition 2. Suppose that k
and m unstable poles are assigned to loops 1 and 2
respectively, i.e.
n3(s)n4(s)
q~'(s) - k+2
de(s ) ill=3(s - Pi)
ns(s)
q~(s) = k+..+e
d3(s) 1-I (s-Pi)
FIG. 2. Decentralized internal model control scheme. i=k+3
Technical Communique 421
where Re {Pl} > 0 for i = 3 - k + m + 2 and dz(s), d3(s ) are (3) Loop 1 is stable and loop 2 is unstable with one unstable
polynomials with only LHP-zeros. n3(s ), n4(s ) and ns(s ) are pole. q';(s) is of (15) and
the polynomials such that
1 + 0.11s
q~(s)=-250 _ l + E s E>0 (20)
n3(s) n4(s) and q~(s)
k+2 ' d2($ ) num(A"(s))=O.11~s2+(E--O.11)s+58.5. (21)
II (s-pl)
i=3
The necessary and sufficient condition for (21) having
are proper, q~'(s), q~(s) are designed such that A"(s) of (7) LHP-zeros only is
satisfies Proposition 1. But then the next design also
stabilizes the plant E > 0.11. (22)
..... n4(s) It is interesting to see the physical interpretation of the
ql(S ) -- dz(s ) stability constraint, (22). From (2), (20) the feedback
controller of loop 2 is given by
n3(s) . q~(s)
q~(s ) = k +2
1 ,, -27.5/ 1 \
[I (s-Pi) sC2(S) = ~ ~1 + 0 . ~ s ) " (23)
i=3
because q'~(s)q~(s) = q~'(s)qT(s ) and A"(s) = A"(s). However Therefore the feedback gain is limited by (22). It should be
this contradicts either (2) or (3) of Proposition 2. Hence (2) tuned below 27.5/0.11 = 2 5 0 to guarantee the stability of
of Proposition 2 is false and one has to stabilize both loops C(s).
for achieving decentralized stabilization. []
5. Conclusion
The next example illustrates Theorem 1. A new decentralized stability condition related to R G A
and NI without requiring the stabilization of the individual
Example 1. loops is given. This result should be viewed as a
1 1 generalization of previous work on R G A and NI. The result
G(s) 1+~.11s[0.85[ 0.28 ] is limited to 2 x 2 plants under multi-loop SISO control and
0.0041
should hopefully motivate further development for larger
It is checked that NI < 0 and three different designs are systems.
discussed.
(1) Loops 1 and 2 are stable; for example, choose Acknowledgements--The authors would like to thank Prof.
M. Morari for his helpful comments on the parametrization
q,(s) = 1 (15) of stabilizing controllers. The financial support of National
+ 0.11s Science Foundation is gratefully acknowledged.
q2(s)=25011+es e>O (16)
References
Then the numerator of A(s) is I. Bristol, E. H. (1966). On a new measure of interaction
for multivariable process control. IEEE Tram. Aut.
Hum (A(s)) = Es 2 + (9.09E + 1)s - 531.818 (17) Control, AC-4, 133-134.
It is obvious that (17) has a RHP-zero for any positive e 2. Grosdidier, P., M. Morari and B. R. Holt (1985).
and G(s) can not be stabilized. Closed-loop properties from steady-state gain informa-
(2) Loop 1 is stable and loop 2 is unstable with two unstable tion. Ind, Engng Chem. Fund., 24, 221-235.
poles, q~(s) is the same as (15) and 3. Manousiouthakis, V. (1989). On the parametrization of
all decentralized stabilizing controllers. Proc. ACC,
1 + 0.11s Pittsburgh, PA, pp. 2108-2111.
q6(s)=250(_l+es)(_l+¢s) e,q~>O. (18)
4. Niederlinski, A. (1971). A heuristic approach to the
It is routine to calculate that design of linear multivariable interacting control systems.
Automatica, 7, 691-701.
Hum (A'(s)) = 0. lle~s 3 + (t 0 - O.11e - 0.11~)s 2 5. Youla, D. C., H. A. Jabr and J. J. Bongiorno (1976).
- (c + ~ - 0. l l ) s - 58.5. (19) Modern W i e n e r - H o p f design of optimal controllers---Part
II. The multivariable case. IEEE Tram. Aut. Control,
Again, G(s) can not be stabilized by any positive • and tp. AC-21, 319-338.