Sie sind auf Seite 1von 3

IEEE TRANSACTIONS ON AUTOMATIC CONTROL, VOL. 46, NO.

11, NOVEMBER 2001 1797

the product control structure on L is also locally definable. Let s 2 L, conflict with each other. We also applied this result to the framework
we derive the control technology 6c by Theorem 3 as established in [2]. It demonstrated that the uncontrollable event set of
6c (s) = f[6 0 (61 0 1 ) [ (62 0 2 )] \ EligL (s) j
the composed system is just the union of the uncontrollable event sets
of two local sysems.
i  6ic ; i 2 f1; 2g; 1 \ (62 0 2 ) \ EligL (s) = ; There still remain some problems in the research of hierarchical and
2 \ (61 0 1 ) \ EligL (s) = ;g: decentralized DES control theory. For example, in [1], a hierarchical

Since 6i 0 6ic  6i 0 i , it follows 1 \ (62 0 62c ) \ EligL (s) =


DES is control consistent if and only if the system satisfies a kernel

;; 2 \ (61 0 61c ) \ EligL (s) = ;, i.e.,


condition. What does this kernel condition actually mean to the high
level system, if the high level is standard? And what if the high level
1  61c 0 (62 0 62c ) \ EligL (s) is locally definable? In the decentralized DES control theory, if there
is a hierarchical structure defined on a local system, say L2 , then such
2  62c 0 (61 0 61c ) \ EligL (s): (12)
hierarchical structure can be extended to the global system L. Now,
By Theorem 3, for all  2 6c (s); assume the hierarchical DES L2 is control consistent, then under what
 = [6 0 (61 0 1 ) [ (62 0 2 )] \ EligL (s)
conditions will the global hierarchical DES be also control consistent.
These will be studied in the future.
for some i satisfying (12). Thus
  [(61c 0 62 ) [ (61c \ 62c ) [ (62c 0 61 )] \ EligL (s) ACKNOWLEDGMENT

Let 0  [(61c 0 62 ) [ (61c \ 62c ) [ (62c 0 61 )] \ EligL (s); 01 =


The authors gratefully thank W. M. Wonham and K. C. Wong for
0 \ 61c 2 61c (p1 (s)) and 02 = 0 \ 62c 2 62c (p2 (s)). Since
their helpful advice and encouragement. The authors also sincerely
thank the reviewers for their valuable comments and suggestions.
6 0 (61 \ 62c 0 61c )
= 61c [ (62 0 61 ) [ (61 0 62c ) [ (62 0 62c ) REFERENCES
 [(61c 0 62 ) [ (61c \ 62c ) [ (62c 0 61 )] \ EligL (s) [1] K. C. Wong, Discrete-Events Control Architecture: An Algebraic Ap-
0 proach, Systems Control Group, Department of Electrical and Com-
puter Engineering, Univ. Toronto, Tech. Rep. 9407, 1994.
[2] P. J. Ramadge and W. M. Wonham, Supervisory of discrete event pro-
we have cesses, in Proc. 21st Conf. Decision Control, 1982, pp. 12281229.
02 \ EligL (s) \ [61 0 01 \ EligL (s)]
= (61 \ 62c 0 61c ) \ 0
= ;:
Likewise, we have 01 \ EligL (s) \ [62 0 02 \ EligL (s)] = ;. With
Plant With Integrator: An Example of Reset Control
0  [(61c 0 62 ) [ (61c \ 62c ) [ (62c 0 61 )] \ EligL (s): Overcoming Limitations of Linear Feedback
By direct calculation, we get O. Beker, C. V. Hollot, and Y. Chait
0 = 6 0 [(61 0 01 ) [ (62 0 02 )]
or, the control technology 6c of the composing system is AbstractThe purpose of this note is twofold. First, to give conditions
under which linear feedback control of a plant containing integrator must
6c (s) = P ([(61c 0 62 ) [ (61c \ 62c ) overshoot. Second, to give an example of reset control that does not over-
[ (62c 0 61 )] \ EligL (s)) shoot under such constraints.

where s 2 L. So, we can also control the global system L by desig-


Index TermsIntegrators, performance limitations, reset control.

nating a subset of the event alphabet 6con as controllable event labels


and the remaining event labels 6unc = 6 0 6con as uncontrollable, I. INTRODUCTION
where
Reset control was introduced by Clegg [1] as a means to overcome
6con = (61c 0 62 ) [ (61c \ 62c ) [ (62c 0 61 ) the limitations of linear feedback. A reset controller is a linear system
= 6 0 (61 0 61c ) [ (62 0 62c ): whose states are reset to zero when its input equals zero. Clegg ap-
plied this concept to an integrator element (the Clegg integrator) where
It is easy to derive that the uncontrollable event-set is 6unc = 61u [
62u , which is the union of two local uncontrollable event labels.
Manuscript received July 13, 2000; revised January 3, 2001. Recommended
by Associate Editor S. Hara. This material is based upon work supported by the
IV. CONCLUSION National Science Foundation under Grant CMS-9800612.
O. Beker was with the Electrical and Computer Engineering Department,
In this note, we investigated the property of locally definable on the
product control structure. We proved that if both Ci are standard and
University of Massachusetts, Amherst, MA 01003 USA. He is now with Maxtor
Corporation, Shrewsbury, MA 01545 USA.
locally definable control structures, so is their product C1 k C2 . And C. V. Hollot is with the Electrical and Computer Engineering Depart-
the control technologies of composed system L are further discussed. ment, University of Massachusetts, Amherst, MA 01003 USA (e-mail:
We derived the control technology associated with C1 k C2 and showed
hollot@ecs.umass.edu).
Y. Chait is with the Mechanical and Industrial Engineering Department, Uni-
that the global eligible event-set of a controllable language can be ex- versity of Massachusetts, Amherst, MA 01003 USA.
pressed by the union of two local control options, which should not Publisher Item Identifier S 0018-9286(01)10355-7.

00189286/01$10.00 2001 IEEE


1798 IEEE TRANSACTIONS ON AUTOMATIC CONTROL, VOL. 46, NO. 11, NOVEMBER 2001

Fig. 1. Linear feedback control system. Fig. 2. Reset control of an integrator using a first-order reset element.

the rationale for improved feedback performance came from this ele- Example: Consider the linear feedback system in Fig. 1 where the
ments favorable describing function. Krishnan and Horowitz [2] in- plant P (s) is simply an integrator. In addition to closed-loop stability,
corporated the Clegg integrator into a quantitative design procedure. In suppose the design objectives are the following:
[3], these ideas were advanced by resetting a first-order lag filter, the
i) steady-state error no greater than one when tracking a unit-ramp
so-called first-order reset element. Recently, there has been renewed at-
input;
tention to this topic from both a theoretic and applications viewpoint;
ii) rise time greater than 2 s when tracking a unit-step;
e.g., see [4][11]. One missing element in this work is a concrete ex-
iii) no overshoot in the step response.
ample showing that reset control meets control system objectives that
To meet the error specification on the ramp response, this linear feed-
back system must have velocity error constant Kv  1. Since tr >
are unattainable over all linear controllers. This paper provides such
2  (2=Kv ), the Proposition indicates that no stabilizing C (s) exists
an example. We note that switching control, a nonlinear scheme analo-
gous to reset, exhibits similar advantage as claimed in [12]. The paper
to meet all the above objectives. In the next section we show that a reset
is organized as follows: In the next section, we introduce an overshoot
control system can meet these specifications.
limitation on linear feedback systems arising when the loop contains
an integrator. While new, this result is an immediate consequence of
the time-domain limitations introduced in [13]. In Section III, we give III. RESET CONTROL MEETS EXAMPLEs SPECIFICATIONS
an example showing that reset control does not suffer this limitation. We reconsider the previous example using reset control (see Fig. 2)
In Section IV, we conclude. and use a first-order reset element, see [3], described by

II. OVERSHOOT IN A LINEAR FEEDBACK SYSTEM CONTAINING u_ (t) = 0u(t) + e(t); e(t) 6= 0
INTEGRATOR u(t+ ) = 0; e(t) = 0:
Consider the standard linear feedback control system in Fig. 1 where In the following, we let base-linear system refer to the reset control
the plant P (s) contains an integrator. Assume that C (s) stabilizes. In system in the absence of resetting. To address the preceding design
[13] it was shown that the tracking error e due to a unit-step input sat- objectives we first compute the tracking error of this base-linear system
isfies to a unit-ramp
1 1 p p
e(t) dt = 2 3 00:5t 3 
0 Kv eramp (t) = 1 + e sin t0 :
3 2 3
1
where the velocity constant Kv is defined by Kv = Since eramp (t) 6= 0, the reset control system never resets and its
lims!0 sP (s)C (s). Alone, this constraint does not imply overshoot response is equivalent to the base-linear response. Kv = 1 for the
in the step response y ; i.e., y (t)  1 for some t > 0. However, base-linear system guarantees a steady-state error less than 1. The error
introduction of an additional, sufficiently stringent time-domain of the base-linear control system to a unit-step input is
bandwidth constraint will. To see this, consider the notion of rise time p p p
tr introduced in [13]: 2 3 00:5t 3 3 
estep (t) =
3
e sin
2
t 0 sin 2
t0
3
:
t p
tr = sup T : y(t)  ; t 2 [0; T ] :
T T This error first goes to zero at t = (4 3=9)  2:42 s. Thus, the first
The following result is immediate. reset time is approximately 2.42 s. The error of the reset control system
Proposition: If tr > (2=Kv ); i.e., the rise time is sufficiently slow, remains zero therafter since its response is deadbeat. This occurs since
the unit-step response y (t) overshoots. (u; y ) = (0; r) is an equilibrium point. The rise time tr is determined
1
from the tangency of y (t) = 1 0 estep (t) with y(t) = mt. Indeed,
Proof: Clearly
1 if (1=m) < 2:42 and these curves are tangent, then tr = (1=m). We
1 thus seek a pair (t; m) satisfying
= e(t) dt
Kv 0
t
t 1 y(t) = y(t)
 10
tr
dt + e(t) dt
0 and
1
t

tr y_ (t) = y_ (t):


= + e(t) dt:
2
This occurs for m = (1=2:35). Since 2:35 < 2:42 we conclude
t

Thus tr = 2:35 s. This reset control system thus satisfies the second de-
1 1
0 t2 :
sign constraint. The overshoot objective is met since the step response
e(t) dt  r

t Kv is deadbeat. Consequently, this reset control system meets objectives


not attainable using linear feedback control. We confirm this analysis
Since tr > (2=Kv ), then e(t) < 0 (and hence y (t) > 1) for some with simulation results shown in Figs. 3 and 4. Finally, one can show
t 2 (tr ; 1). the unforced reset control system to be asymptotically stable and to
IEEE TRANSACTIONS ON AUTOMATIC CONTROL, VOL. 46, NO. 11, NOVEMBER 2001 1799

[4] R. T. Bupp, D. S. Bernstein, V. Chellaboina, and W. Haddad, Resetting


virtual absorbers for vibration control, in Proc. Amer. Control Conf.,
Albuquerque, NM, 1997, pp. 26472651.
[5] J. E. Bobrow, F. Jabbari, and K. Thai, An active truss element and
control law for vibration suppression, Smart Mater.Struct., vol. 4, pp.
264269, 1995.
[6] H. Hu, Y. Zheng, Y. Chait, and C. V. Hollot, On the zero-input sta-
bility of control systems having clegg integrators, in Proc. Amer. Con-
trol Conf., Albuquerque, NM, 1997, pp. 408410.
[7] O. Beker, C. V. Hollot, Q. Chen, and Y. Chait, Stability of a reset control
system under constant inputs, in Proc. Amer. Control Conf., San Diego,
CA, 1999, pp. 30443045.
[8] Y. Zheng, Y. Chait, C. V. Hollot, M. Steinbuch, and M. Norg, Experi-
mental demonstration of reset control, IFAC J. Control Eng. Practice,
vol. 8, no. 2, pp. 113120, 2000.
[9] Q. Chen, C. V. Hollot, and Y. Chait, Stability and asymptotic perfor-
mance analysis of a class of reset control systems, Proc. 39th IEEE
Conf. Decision Control, pp. 251256, 2000.
[10] Q. Chen, C. V. Hollot, Y. Chait, and O. Beker, On reset control systems
with second-order plants, in Proc. Amer. Control Conf., Chicago, IL,
2000, pp. 205209.
[11] W. M. Haddad, V. Chellaboina, and N. A. Kablar, Active control of
Fig. 3. Tracking-error e to a unit-ramp input for the reset control system. combustion instabilities via hybrid resetting controllers, in Proc. Amer.
Control Conf., Chicago, IL, 2000, pp. 23782382.
[12] A. Feuer, G. C. Goodwin, and M. Salgado, Potential benefits of hybrid
control for linear time invariant plants, in Proc. Amer. Control Conf.,
Albuquerque, NM, 1997, pp. 27902794.
[13] R. H. Middleton, Trade-offs in linear control system design, Auto-
matica, vol. 27, no. 2, pp. 281292, 1991.
[14] C. V. Hollot, O. Beker, Y. Chait, and Q. Chen, On establishing classic
performance measures for reset control systems, in Perspectives in Ro-
bust Control, Lecture Notes in Control and Information Sciences 268, S.
O. R. Moheimani, Ed. New York: Springer-Verlag, 2001, pp. 123147.

On Quadratic Stability of Systems With Structured


Uncertainty
Wei-Yong Yan and James Lam

AbstractThis note considers the problem of stability robustness


Fig. 4. Output response y to a unit-step input for the reset control system with respect to a class of nonlinear time-varying perturbations which
(dotted line shows rise time constraint). are bounded in a component-wise rather than aggregated manner. A
family of robustness bounds is parameterized in terms of a nonsingular
symmetric matrix. It is shown that the problem of computing the largest
track step inputs with zero steady-state error; see [7], [9], and [14] for robustness bound over the set of nonsingular symmetric matrices can be
more details. approximated by a smooth minimization problem over a compact set.
A convergent algorithm for computing an optimal robustness bound is
proposed in the form of a gradient flow.
IV. CONCLUSION
Index TermsLinear systems, optimization, quadratic stability, stability
The main contribution of this note is an example of control specifica- robustness.
tions that can be achieved by reset control and not by linear feedback.
This does not imply that reset control is superior; rather, that reset con-
trol has a different set of performance limitations. Such differences can I. INTRODUCTION
be exploited in specific control applications as demonstrated in [4], [5], Linear systems are often subject to time-varying nonlinear perturba-
[8], and [11]. tions including parametric uncertainties. The issue of stability robust-

REFERENCES
Manuscript received November 30, 1999; revised July 25, 2000 and February
[1] J. C. Clegg, A nonlinear integrator for servomechanisms, Trans. AIEE, 26, 2001. Recommended by Associate Editor S. Hara. This work was supported
Part II, Appl. Ind., vol. 77, pp. 4142, 1958. in part an ARC research grant, a HKU CRCG grant, and the William Mong
[2] K. R. Krishnan and I. M. Horowitz, Synthesis of a nonlinear feedback Visiting Fellowship.
system with significant plant-ignorance for predescribed system toler- W.-Y. Yan is with the School of Electrical and Computer Engineering, Curtin
ances, Int. J. Control, vol. 19, no. 4, pp. 689706, 1974. University of Technology, Perth, Australia.
[3] I. Horowitz and P. Rosenbaum, Non-linear design for cost of feedback J. Lam is with the Department of Mechanical Engineering, University of
reduction in systems with large parameter uncertainty, Int. J. Control, Hong Kong, Pokfulam Road, Hong Kong.
vol. 24, no. 6, pp. 9771001, 1975. Publisher Item Identifier S 0018-9286(01)10356-9.

00189286/01$10.00 2001 IEEE