BUMPLESS SWITCHING CONTROLLERS. William A. Wolovich and Alan B. Arehart 1. December 27, Abstract

Similar documents
Multivariable MRAC with State Feedback for Output Tracking

CONTROL DESIGN FOR SET POINT TRACKING

Fall 線性系統 Linear Systems. Chapter 08 State Feedback & State Estimators (SISO) Feng-Li Lian. NTU-EE Sep07 Jan08

PARAMETERIZATION OF STATE FEEDBACK GAINS FOR POLE PLACEMENT

MANY adaptive control methods rely on parameter estimation

AN OPTIMIZATION-BASED APPROACH FOR QUASI-NONINTERACTING CONTROL. Jose M. Araujo, Alexandre C. Castro and Eduardo T. F. Santos

Zeros and zero dynamics

and the nite horizon cost index with the nite terminal weighting matrix F > : N?1 X J(z r ; u; w) = [z(n)? z r (N)] T F [z(n)? z r (N)] + t= [kz? z r

Discussion on: Measurable signal decoupling with dynamic feedforward compensation and unknown-input observation for systems with direct feedthrough

WEAK STRUCTURE AT INFINITY AND ROW-BY-ROW DECOUPLING FOR LINEAR DELAY SYSTEMS

Performance assessment of MIMO systems under partial information

Chapter 3. LQ, LQG and Control System Design. Dutch Institute of Systems and Control

Module 03 Linear Systems Theory: Necessary Background

Homework 6 Solutions. Solution. Note {e t, te t, t 2 e t, e 2t } is linearly independent. If β = {e t, te t, t 2 e t, e 2t }, then

ECE 388 Automatic Control

Design Methods for Control Systems

QUALITATIVE CONTROLLABILITY AND UNCONTROLLABILITY BY A SINGLE ENTRY

Multivariable ARMA Systems Making a Polynomial Matrix Proper

1 Introduction 198; Dugard et al, 198; Dugard et al, 198) A delay matrix in such a lower triangular form is called an interactor matrix, and almost co

Chapter 30 Minimality and Stability of Interconnected Systems 30.1 Introduction: Relating I/O and State-Space Properties We have already seen in Chapt

STAT 309: MATHEMATICAL COMPUTATIONS I FALL 2018 LECTURE 13

Intro. Computer Control Systems: F8

A Simple Derivation of Right Interactor for Tall Transfer Function Matrices and its Application to Inner-Outer Factorization Continuous-Time Case

The Rationale for Second Level Adaptation

Full State Feedback for State Space Approach

Control Systems Design

11 Three problems on the decidability and complexity of stability

16.31 Fall 2005 Lecture Presentation Mon 31-Oct-05 ver 1.1

Chapter 9 Observers, Model-based Controllers 9. Introduction In here we deal with the general case where only a subset of the states, or linear combin

ẋ n = f n (x 1,...,x n,u 1,...,u m ) (5) y 1 = g 1 (x 1,...,x n,u 1,...,u m ) (6) y p = g p (x 1,...,x n,u 1,...,u m ) (7)

arxiv: v1 [cs.sy] 2 Apr 2019

I = i 0,

Group Theory. 1. Show that Φ maps a conjugacy class of G into a conjugacy class of G.

Decentralized control with input saturation

ON CHATTERING-FREE DISCRETE-TIME SLIDING MODE CONTROL DESIGN. Seung-Hi Lee

Cayley-Hamilton Theorem

Edge Theorem for Multivariable Systems 1

1 Linear Algebra Problems

IMPULSIVE CONTROL OF DISCRETE-TIME NETWORKED SYSTEMS WITH COMMUNICATION DELAYS. Shumei Mu, Tianguang Chu, and Long Wang

Supervisory Control of Petri Nets with. Uncontrollable/Unobservable Transitions. John O. Moody and Panos J. Antsaklis

Linear Quadratic Gausssian Control Design with Loop Transfer Recovery

Reachability of a class of discrete-time positive switched systems

Adaptive and Robust Controls of Uncertain Systems With Nonlinear Parameterization

The norms can also be characterized in terms of Riccati inequalities.

Algebra Homework, Edition 2 9 September 2010

Krylov Techniques for Model Reduction of Second-Order Systems

Recent Advances in Positive Systems: The Servomechanism Problem

MODERN CONTROL DESIGN

Observability. Dynamic Systems. Lecture 2 Observability. Observability, continuous time: Observability, discrete time: = h (2) (x, u, u)

Lyapunov Stability of Linear Predictor Feedback for Distributed Input Delays

EL2520 Control Theory and Practice

STABILITY OF INVARIANT SUBSPACES OF COMMUTING MATRICES We obtain some further results for pairs of commuting matrices. We show that a pair of commutin

Control for Coordination of Linear Systems

Controlo Switched Systems: Mixing Logic with Differential Equations. João P. Hespanha. University of California at Santa Barbara.

Canonical lossless state-space systems: staircase forms and the Schur algorithm

INRIA Rocquencourt, Le Chesnay Cedex (France) y Dept. of Mathematics, North Carolina State University, Raleigh NC USA

Zero controllability in discrete-time structured systems

6.241 Dynamic Systems and Control

Modeling and Analysis of Dynamic Systems

Robust Control 2 Controllability, Observability & Transfer Functions

Model reduction for linear systems by balancing

Switched Systems: Mixing Logic with Differential Equations

ELE/MCE 503 Linear Algebra Facts Fall 2018

ANALYTICAL MATHEMATICS FOR APPLICATIONS 2018 LECTURE NOTES 3

Spectral factorization and H 2 -model following

DESIGN OF LINEAR STATE FEEDBACK CONTROL LAWS

Basic Concepts in Linear Algebra

What is A + B? What is A B? What is AB? What is BA? What is A 2? and B = QUESTION 2. What is the reduced row echelon matrix of A =

ACI-matrices all of whose completions have the same rank

Math 240 Calculus III

MAT 1302B Mathematical Methods II

H 2 Suboptimal Estimation and Control for Nonnegative

SUPPLEMENT TO CHAPTERS VII/VIII

Control Design of a Distributed Parameter. Fixed-Bed Reactor

Row and Column Representatives in Qualitative Analysis of Arbitrary Switching Positive Systems

State Feedback and State Estimators Linear System Theory and Design, Chapter 8.

Module 09 From s-domain to time-domain From ODEs, TFs to State-Space Modern Control

Structural Properties of LTI Singular Systems under Output Feedback

Matrix Operations: Determinant

Matrices and Vectors

VII Selected Topics. 28 Matrix Operations

Stabilization, Pole Placement, and Regular Implementability

Homework Set 5 Solutions

1 Introduction. 2 Determining what the J i blocks look like. December 6, 2006

SUCCESSIVE POLE SHIFTING USING SAMPLED-DATA LQ REGULATORS. Sigeru Omatu

CANONICAL LOSSLESS STATE-SPACE SYSTEMS: STAIRCASE FORMS AND THE SCHUR ALGORITHM

State feedback gain scheduling for linear systems with time-varying parameters

CANONICAL FORMS FOR LINEAR TRANSFORMATIONS AND MATRICES. D. Katz

POSITIVE PARTIAL REALIZATION PROBLEM FOR LINEAR DISCRETE TIME SYSTEMS

Semi-global Robust Output Regulation for a Class of Nonlinear Systems Using Output Feedback

Controllability, Observability, Full State Feedback, Observer Based Control

REMARKS ON THE TIME-OPTIMAL CONTROL OF A CLASS OF HAMILTONIAN SYSTEMS. Eduardo D. Sontag. SYCON - Rutgers Center for Systems and Control

Regular Sparse Crossbar Concentrators

Central Groupoids, Central Digraphs, and Zero-One Matrices A Satisfying A 2 = J

CONTROL SYSTEMS ANALYSIS VIA BLIND SOURCE DECONVOLUTION. Kenji Sugimoto and Yoshito Kikkawa

On linear quadratic optimal control of linear time-varying singular systems

Fuzzy control of a class of multivariable nonlinear systems subject to parameter uncertainties: model reference approach

ACM/CMS 107 Linear Analysis & Applications Fall 2016 Assignment 4: Linear ODEs and Control Theory Due: 5th December 2016

ECEN 420 LINEAR CONTROL SYSTEMS. Lecture 6 Mathematical Representation of Physical Systems II 1/67

Stability of interval positive continuous-time linear systems

Transcription:

BUMPLESS SWITCHING CONTROLLERS William A. Wolovich and Alan B. Arehart 1 December 7, 1995 Abstract This paper outlines the design of bumpless switching controllers that can be used to stabilize MIMO plants which are both switching stabilizable and switching detectable. A primary feature of such controllers is that their order is no greater than that of the plant. 1 Introduction Consider a linear, time-invariant MIMO dynamical system or plant dened by the state-space representation: _x(t) = Ax(t) + Bu(t); y(t) = Cx(t); (1:1) with A < nn, B < nm and C < pn, which is characterized by the open-loop transfer matrix T yu (s) = C(sI? A)?1 B Such plants will be called switching systems if one or more of the elements or parameters which comprise the switching matrices A, B and C can asynchronously switch from one xed value to another. The (nite) set of all possible xed plants, or fa; B; Cg triples which belong to a particular switching system, will be called its switching group. In particular, if A i (for i = 1; ; : : :; ), B j (for j = 1; ; : : :; ) and C k (for k = 1; ; : : :; ) denote all possible xed values of A, B and C in a particular switching system, then the ( ) triples fa i ; B j ; C k g will dene its switching group. To illustrate the preceding, a sudden failure of (say) only input u q (t) could be modeled by a switching system where all of the q-th column elements of B instantaneously switch to. In such a case, two xed plants fa 1 ; B 1 ; C 1 g and fa 1 ; B ; C 1 g would dene its switching group. More generally, if a switching system were dened by the condition that any (but only one) of its m > 1 inputs could fail, its switching group would consist of exactly m + 1 xed plants, with both A = A 1 and C = C 1 xed, and B = B j, for j = 1; ; : : :; m + 1 =. 91 1 Laboratory for Engineering Man/Machine Systems, Box D, Brown University, Providence, RI 1

A switching system will be called bumpless if its output y(t) remains a continuous function of time when u(t) is continuous, despite all possible parameter switches. Clearly, a switching system will be bumpless if the output matrix C remains xed, and only the parameters of A and B (which appear before the state integrators) can switch values. A switching system will be called switching controllable if all members of its switching group are (completely state) controllable. Switching observability, switching stabilizability and switching detectability are dened in an analogous manner. It is of interest to note that xed controllers sometimes can be found that stabilize more than one member of a switching group. Such would be the case, for example, if a single xed controller were to simultaneously stabilize two or more xed plants[1][], all of which belong to the switching group. Morse and Pait have done extensive work on switching controllers, which are switching systems used as feedback controllers [5] [4] [6] [7]. In [6], they discuss the design of bumpless switching controllers for MIMO plants, noting a need to reduce the order of such controllers. This paper will address this issue. n-th Order Non-Bumpless Switching Controllers If a switching system is both switching stabilizable and switching detectable, then it is well known that ( ) and ( ) switching gain matrices, namely G ij and K ik, can be determined which arbitrarily assign all of the controllable eigenvalues of A i + B j G ij and all of the observable eigenvalues of A i + K ik C k for all switching group members. In such cases, both A i +B j G ij and A i +K ik C k can be stabilized in the sense that all of their eigenvalues will lie in the half-plane Re(s) <. Therefore, a switching observer, as dened by the state-space representation: _~x(t) = (A i + K ik C k )~x(t) + B j u(t)? K ik y(t); u(t) = G ij ~x(t) + r(t); with r(t) an external reference input, would then imply a (generally non-bumpless) switching compensator, as dened by the transfer matrix T uy (s) =?G ij (si? A i? B j G ij? K ik C k )?1 K ik ; (:1) and a corresponding family of closed-loop systems dened by the transfer matrices T yr (s) = C k (si? A i? B j G ij )?1 B j jsi? A i? K ik C k j jsi? A i? K ik C k j = C k(si? A i? B j G ij )?1 B j

Switching Conversion Matrices If all ( ) of the (m n) switching gain matrices G ij are of full rank m, then for any xed (m n) gain matrix F, also of full rank m, ( ) non-unique, but nonsingular (n n) switching conversion matrices P ij can be determined such that G ij = F P ij (:1) One way to determine such P ij is to rst determine any m linearly independent columns of G ij and F, which will be dened as G mij and F m, respectively. The permutation matrices P Gij and P F could then be used to reorder the columns of G ij and F so that G ij P Gij = [ G mij G rij ] and F P F = [ F m F r ] A subsequent choice of F?1 m M ij = G mij F m?1 [G rij? F r ] I (:) would then imply that G ij P Gij = F P F M ij, or that G ij = F P ij for P ij = P F M ij P?1 Gij (:) Note that if F = [I ] = [F m F r ], then P F = I. Moreover, if the rst m columns of all G ij = [G mij G rij ] are linearly independent, then P Gij = I as well. In such cases, (.) and (.) will imply that Gmij G P ij = M ij = rij (:4) I 4 n-th Order Bumpless Switching Controllers We rst observe that F P (si? A)?1 B = F (si? P AP?1 )?1 P B Therefore, when a switching permutation matrix P ij is determined such that (.1) holds, the switching compensator dened by (.1) also can be dened by T uy (s) =?F (si? P ij [A i + B j F P ij + K ik C k ]P?1 ij )?1 P ij K ik ; (4:1) Nonsingular matrices which have a single nonzero entry of 1 in each row and column.

and implemented in the \standard" observer conguration with T ur (s) = F (si? P ij [A i + B j F P ij + K ik C k ]P?1 ij P )?1 ij B j + I This \replacement" of G ij by F P ij, with F xed, converts a generally non-bumpless compensator dened by (.1) into the bumpless compensator dened by ( 4.1). An Example of Integrity To illustrate the preceding, consider a two-input state-space system dened by the matrix pair? 1 A = 4?1 1 5 and B = [ b 1 b ] = 4 1 5 ; 1?1 5 4 1 which is (completely state) controllable using either input alone. We assume that either of the two inputs can fail suddenly with a resultant, instantaneous switch of B = B 1 to either B = [ b ] = 4 5 or B = [ b 1 ] = 4 1 5 ; 1 4 the (= ) values of B j (with A = A 1 xed) which dene the switching group in this case. We next determine (see the Appendix) that if the switching gain matrix?1?1 G 11 = = [ G?17 8? m11 G r11 ] switches to G 1 = G 1 =?8 8?11 = [ G? 11?6 m1 G r1 ] = [ G m1 G r1 ] ; if either input fails, then the eigenvalues of A 1 +B 1 G 11, A 1 +B G 1 and A 1 +B G 1 will remain at?1 and?1 j in the complex plane. Moreover, G m11 and G m1 = G m1 have full rank in this case. Therefore, if we choose (a xed) 1 F = = [ F 1 m F r ] = [ I ] 4

then (.4) will imply that in the event of a failure of either input, a switch of to P 11 = M 11 =?1?1 G r11 = 4?17 8? 5 I 1 Gm11?8 8?11 P 1 = P 1 = M 1 = M 1 = 4? 11?6 5 ; 1 and a corresponding switch of B 1 to either B or B will insure input integrity[]; i.e. a closed-loop system which remains stable despite the failure of either input. 5 General Compensators We next show that any two strictly proper loop compensators of minimal dynamic order n 1 and n, which have the same number m of (linearly independent) outputs always can be switched in a bumpless manner. In particular, suppose T uy1 (s) = C 1 (si? A 1 )?1 B 1 and T uy (s) = C (si? A )?1 B dene the transfer matrices of the compensators in terms of their state-space representations. If we assume for convenience that n 1 n, then n 1? n = ~n uncon- def trollable and unobservable (stable) modes can be added to the second compensator by appropriately increasing the dimensions of A, B and C. In particular, ~n zero columns (rows) can be added to to C (B ) and a stable (~n ~n) lower right \block" can be added to A to dene C ~ ( B ~ ) and A ~, respectively, which would imply a state-space system of order n 1 with transfer matrix T uy (s). A nonsingular matrix P could then be determined, as in (.), so that C ~ = C 1 P, thereby implying that T uy (s) = ~ C (si? ~ A )?1 ~ B = C 1 (si? P ~ A P?1 )?1 P ~ B Since the output matrix of both compensators would be the xed matrix C 1, either compensator could be switched to the other while maintaining a continuous output. APPENDIX By switching A 1 and B 1 to P ~ AP?1 and P ~ B, respectively, or vice-versa. 5

Consider any controllable system dened by ( 1.1) with B = [ b 1 b : : : b m ] of full (column) rank m < n. If we lexicographically reorder the rst n linearly independent columns of the controllability matrix C = [B AB : : : A n?1 B], as in [9], we will dene the (n n) nonsingular matrix C nn = [b 1 Ab 1 : : : A d 1?1 b 1 b Ab : : :A d?1 b : : :A dm?1 b m ]; the m controllability indices d i, and the (m) integers k = P k d 1 k, so that 1 = d 1, = d 1 + d, etc., with m = d 1 + d + : : : d m = n. If the (m) row vectors q k are then dened as the k rows of C nn,?1 def ^B m = 6 4 q 1 A d 1?1 B q A d?1 B. q m A dm?1 B will be a nonsingular, upper right triangular matrix, with all diagonal elements equal to 1, so that j ^Bm j = 1. If we choose m arbitrary stable monic polynomials i (s), each of degree d i, and dene the state feedback gain matrix G =? ^B?1 m 6 4 q 1 1 (A) q (A). q m m (A) it then follows[8] that the (n) eigenvalues of A + BG will correspond to the roots of (s) def = Q m 1 i(s). It might be noted that in the single-input (m = 1 and B = b) case, a single vector q would be dened as the last row of C?1. For any stable monic polynomial (s) of degree n, a choice of g =?q(a) 4 would then imply that the (n) eigenvalues of A + bg will correspond to the roots of (s). 7 5 7 5 ; References [1] A. B. Arehart and W. A. Wolovich, \A Nonlinear Programming Procedure and a Necessary Condition for the Simultaneous Stabilization of or More Linear Systems," Proc. IEEE Conf. on Decision and Contr., 1995. 4 Which is known as Ackermann's formula[1]. 6

[] Blondel, Vincent, Simultaneous Stabilization of Linear Systems, Springer-Verlag Lecture Notes in Control and Information Sciences 191, 1994. [] Davison, Edward J., \The Design of Controllers for the Multivariable Robust Servomechanism Problem Using Parameter Optimization Methods," IEEE TAC, Vol. 6 (1), February, 1981. [4] A. S. Morse, \Towards a Unied Theory of Parameter Adaptive Control{Part : Certainty Equivalence and Implicit Tuning," IEEE Trans. Automat. Contr., vol. 7, pp. 15{9, Jan. 199. [5] A. S. Morse, \Control Using Logic-Based Switching," Proc. European Contr. Conf., 1995. [6] A. S. Morse and F. M. Pait, \MIMO Design Models and Internal Regulators for Cyclicly Switched Parameter-Adaptive Control Systems," IEEE Trans. Automat. Contr., vol. 9, pp. 189{1818, Sept. 1994. [7] F. M. Pait and A. S. Morse, \A Cyclic Switching Strategy for Parameter- Adaptive Control," IEEE Trans. Automat. Contr., vol. 9, pp. 117{118, Jun. 1994. [8] J. Wang and Y. Juang, \A New Approach for Computing the State Feedback Gains of Multivariable Systems," IEEE Trans. Automat. Contr., vol. 4, pp. 18{186, Oct. 1995. [9] W. A. Wolovich, Linear Multivariable Systems. Springer, 1974. [1] W. A. Wolovich, Automatic Control Systems: Basic Analysis and Design. Saunders, 1994. 7